Operational evaluation of Do Not Block the Box campaigns in Georgia

GEORGIA DOT RESEARCH PROJECT 14-43
FINAL REPORT
OPERATIONAL EVALUATION OF DO NOT BLOCK THE BOX CAMPAIGNS
IN GEORGIA
OFFICE OF PERFORMANCE-BASED MANAGEMENT AND RESEARCH 15 KENNEDY DRIVE FOREST PARK, GA 30297-2534

GDOT Research Project No. RP 14-43 Final Report
OPERATIONAL EVALUATION OF DO NOT BLOCK THE BOX CAMPAIGNS IN GEORGIA
By Angshuman Guin, Ph.D., Principal Investigator Michael Hunter, Ph.D., co-Principal Investigator Michael Rodgers, Ph.D., co-Principal Investigator Abhilasha Saroj, Graduate Research Assistant Nishu Choudhary, Graduate Research Assistant
Han Gyol Kim, Graduate Research Assistant School of Civil and Environmental Engineering, Georgia Institute of Technology
Contract with
Georgia Department of Transportation
In cooperation with
U.S. Department of Transportation Federal Highway Administration
April 2018
The contents of this report reflect the views of the authors who are responsible for the facts and accuracy of the data presented herein. The contents do not necessarily reflect the official views or policies of the Georgia Department of Transportation or the Federal Highway Administration. This report does not constitute a standard, specification, or regulation.

Operational Evaluation of Do Not Block the Box Campaigns in Georgia

Technical Report Documentation Page

1. Report No.
FHWA-GA-18-1443

2. Government Accession Number

4. Title and Subtitle
Operational Evaluation of Do Not Block the Box Campaigns in Georgia

3. Recipient's Catalog No.
5. Report Date
April 23, 2018
6. Performing Organization Code

7. Author(s)

8. Performing Organization Report

Angshuman Guin, Ph.D.; Michael Hunter, Ph.D.; Michael Rodgers, Ph.D.; No.

Abhilasha Saroj, Nishu Choudhary, Han Gyol Kim

2006Y62

9. Performing Organization Name and Address
Georgia Tech Research Corp. 505 Tenth St. Atlanta, GA 30318
12. Sponsoring Organization Name and Address
Georgia Department of Transportation Office of Performance-Based Management and Research 15 Kennedy Drive Forest Park, GA 30297-2534

10. Work Unit No.:
11. Contract or Grant No.
0013114
13. Type of Report & Period Covered
Final; December 2015February 2018
14. Sponsoring Agency Code

15. Supplementary Notes
Prepared in cooperation with the U.S. Department of Transportation, Federal Highway Administration.

16. Abstract
The overarching objective of this research is to provide the Georgia Department of Transportation (GDOT) and the Perimeter Community Improvement District (PCID) with an evaluation of the operational performance impacts of implementing a "Do Not Block the Box" (DBTB) campaign at selected signalized intersections. For the studied sites, the likelihood, or propensity, of a vehicle to block was measured both before and after the DBTB treatment installation. Several blocking behavior characteristics were seen throughout the analysis. First, the change in propensity with the installation of the DBTB treatment was inconsistent, witnessing both increasing and decreasing blocking rates. However, regardless of an increase or decrease in blocking rate, the aggregated observed propensities at the studied intersections were consistently high in both the before and after treatment conditions. The lowest observed aggregate propensity to block was 55%, with all other time periods above 60%, and with half of the observed periods having a propensity to block of 70%. In addition, there was significant variability in dayto-day blocking opportunities. While the current treatment as a standalone measure did not meaningfully impact blocking behavior, there is significant value in continuing to seek reductions in blocking behavior. Based on the study findings and field observations during the data collection, several recommendations are offered in this report.

17. Key Words

18. Distribution Statement

Do Not Block the Box, Traffic Control, Roadway Safety,

Simulation

19. Security Class (This Report)

20. Security Class (This Page)

21. No of Pages

UNCLASSIFIED

UNCLASSIFIED

157

Form DOT F 1700.7 (8/72) Reproduction of form and completed page is authorized

22. Price

TABLE OF CONTENTS

List of Tables ......................................................................................................... iv

List of Figures ..........................................................................................................v

Executive Summary .............................................................................................. vii

: INTRODUCTION.................................................................................1 1.1 Background ..........................................................................................................2 1.2 Project Objectives ................................................................................................3

: DATA EXTRACTION AND ANALYSIS METHODOLOGY...........5

2.1 Introduction..........................................................................................................5

2.2 Video Data Collection Process ............................................................................6

2.3 Video Data Extraction Software ........................................................................10

2.4 Methodology to Estimate Propensity to Block ..................................................16

2.4.1 2.4.2 2.4.3 2.4.4

Step 1. Initial Review...................................................................................17 Step 2. Blocking Event Data Extraction ......................................................18 Step 3. Day, Time, and Lane Selection........................................................24 Step 4. Block-Source Lane Vehicle Categorization .....................................24

: DATA AND FINDINGS ....................................................................27

3.1 Intersection Selection.........................................................................................27

3.1.1 3.1.2 3.1.3 3.1.4

PCID Before.................................................................................................27 PCID After ...................................................................................................28 GDOT Before...............................................................................................31 GDOT After .................................................................................................32

3.2 Intersection Results ............................................................................................34

3.2.1 3.2.2 3.2.3 3.2.4 3.2.5 3.2.6

Ashford Dunwoody Rd. NE at Ravinia Dr. NE...........................................36 Peachtree Dunwoody Rd. at Lake Hearn Dr. NE ........................................38 Peachtree Dunwoody Rd. at Abernathy Rd. NE..........................................40 Peachtree Dunwoody Rd. at Johnson Ferry Rd. ..........................................41 W. Paces Ferry Rd. NW at I-75 SB On/Off Ramps.....................................43 Clairmont Rd. at I-85 SB, near Sam's Club.................................................44

3.3 Field Result Summary........................................................................................46

3.4 Simulation Study................................................................................................48

i

Operational Evaluation of Do Not Block the Box Campaigns in Georgia

: CONCLUSIONS AND RECOMMENDATIONS .............................49 4.1 Conclusions........................................................................................................49 4.2 Recommendations..............................................................................................51

: REFERENCES....................................................................................55

Appendix A ANALYSIS OF VEHICLE BLOCKING BEHAVIOR ON INTERSECTION PERFORMANCE .............................................................57

A.1 Abstract ..............................................................................................................58

A.2 Introduction........................................................................................................59

A.3 Background ........................................................................................................60

A.4 Methodology ......................................................................................................63

A.4.1 A.4.2 A.4.3 A.4.4 A.4.5

PTV VISSIM Traffic Simulation Software ..............................................63 Network Layout ...........................................................................................64 Implementing Blocking Behavior................................................................64 Dynamic Assignment of Vehicle Type........................................................65 Implementing Blocking Rules .....................................................................67

A.5 Experimental Design..........................................................................................71

A.6 Results................................................................................................................72

A.6.1 Delay ............................................................................................................72 A.6.2 Minor Street Capacity Reduction.................................................................76

A.7 Conclusion .........................................................................................................77

A.8 References..........................................................................................................78

Appendix B STUDY SITES..................................................................................83

Appendix C VIDEO TOOL FOR MANUALLY EXTRACTING COMPLEX TRAFFIC DATA..........................................................................................133
C.1 Abstract ............................................................................................................133
C.2 Introduction......................................................................................................134
C.3 Background ......................................................................................................136
C.4 Multi-video Player Application .......................................................................138
C.4.1 Case Study: Data Extraction for Intersection Blocking Study...................143 C.4.2 Blocking Event Data Extraction ................................................................144 C.4.3 Using GT-MVP's Blocking Event Extraction Module for Data Extraction .................................................................................................................146 C.4.4 Propensity to Block Data Extraction..........................................................147 C.4.5 Using GT-MVP's Propensity Extraction Module for Data Extraction......148 C.4.6 Other Potential Application Cases .............................................................149
Vehicle and Pedestrian Behavior under Varying Crosswalk Treatments. .............151

ii

Operational Evaluation of Do Not Block the Box Campaigns in Georgia C.5 Verification of GT-MVP: Propensity to Block Data Extraction .....................151 C.6 Conclusions......................................................................................................152 C.7 Acknowledgements..........................................................................................154 C.8 References........................................................................................................155
iii

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
LIST OF TABLES
Table 1: Comparison of VideoAnalyzer and GT-MVP Upgrades............................12 Table 2: PCID Before DBTB Installation: Intersections and Video Recording Dates
and Days. ..............................................................................................................28 Table 3: PCID After DBTB Installation: Intersections and Video Recording Dates
and Days. ..............................................................................................................29 Table 4: Additional GDOT Before Intersections and Video Recording Dates and
Days. .....................................................................................................................32 Table 5: Intersections Dropped from the Study. .......................................................33 Table 6: Additional GDOT After Intersections and Video Recording Dates and Days.
..............................................................................................................................34 Table 7: Final Propensity Analysis Intersections and Lanes. ...................................36 Table 8: Ashford Dunwoody Rd. NE at Ravinia Dr. NE..........................................37 Table 9: Peachtree Dunwoody Rd. at Lake Hearn Dr. NE. ......................................39 Table 10: Peachtree Dunwoody Rd. at Abernathy Rd. NE.......................................41 Table 11: Peachtree Dunwoody Rd. at Johnson Ferry Rd. .......................................42 Table 12: W. Paces Ferry Rd. NW at I-75 SB On/Off Ramps. ................................44 Table 13: Clairmont at I-85 SB, near Sam's Club. ...................................................45 Table 14: Results Summary for All Intersections.....................................................47
iv

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
LIST OF FIGURES
Figure 1: Video Data Recording Equipment at 10th Street NW and Williams Street NW Intersection. ....................................................................................................7
Figure 2: Enhancement in Video Quality at Intersection of Peachtree Dunwoody Rd. at Johnson Ferry Rd., Comparing: (a) Standard Definition Vendor Unit, and (b) Georgia Tech's Portable Video Recording Platform. ............................................9
Figure 3: Snapshot of Customized Blocking Event Extraction Sheet. .....................11
Figure 4: Example Timestamp Extraction Using VideoAnalyzer for Vehicle Entering the Intersection Timestamp Taken When Vehicle in Blue Box Crosses the Crosswalk. ............................................................................................................11
Figure 5: Screen Capture of GT-MVP Software Interface. ......................................13
Figure 6: Snapshots of (a) GT-MVP's Blocking Event Data Extraction Window (9), and (b) GT-MVP's Block-Source Lane Categorization Data Extraction Window (9). ........................................................................................................................15
Figure 7: Flow Chart of Methodology Used to Estimate Propensity to Block. ........17
Figure 8: Snapshot of the Spreadsheet Used to Record Flagged Videos in the Initial Review Process.....................................................................................................18
Figure 9: Snapshot of Blocking Event Extraction Sheet for the Intersection of Clairmont Rd. at I-85 SB, near Sam's Club, 04-17-2017 18:11 (Video 1: GP113441_20170417_181134, Video 2: GP083445_20170417_182317). .......19
Figure 10: Lane Movement Numbers Assigned at Peachtree Dunwoody Rd. at Johnson Ferry Rd..................................................................................................20
Figure 11: Vehicles in Red Blocked the Vehicles Marked in Blue (Peachtree Dunwoody Rd. at Abernathy Rd. NE, 10/28/2015, 17:02). .................................21
Figure 12: Example Blocking Event that Does Not Obstruct Traffic Flow (Peachtree Dunwoody Rd. at Johnson Ferry Rd., 10/26/2015, 16:00 1_20151026_160000_0001n0_cam3.avi). ...........................................................22
Figure 13: Blocking Event with No Presence of Demand (Ashford Dunwoody Rd. NE at Ravinia Dr. NE, 10/29/2015, 16:39). .........................................................23
Figure 14: Snapshot: Propensity Extraction Interface of GT-MVP (Peachtree Dunwoody Rd. at Johnson Ferry Rd., 04-27-2016, 16:22 PM GP062213/GP070993) with Extracted Propensity Data Spreadsheet. .................25
v

Operational Evaluation of Do Not Block the Box Campaigns in Georgia Figure 15: Ashford Dunwoody Rd. NE at Ravinia Dr. NE Source: Google Street
View. ....................................................................................................................30 Figure 16: Ashford Dunwoody Rd. NE at Ravinia Dr. NE, Video Screen Capture,
Location 1, 04/21/2016.........................................................................................30 Figure 17: Ashford Dunwoody Rd. NE at Ravinia Dr. NE, Video Screen Capture,
Location 2, 04/21/2016.........................................................................................31
vi

EXECUTIVE SUMMARY
Traffic congestion is often an unfortunate reality. As such, it is necessary to manage congestion, minimizing its impacts. In congestion, when a vehicle enters an intersection that has insufficient space to exit on the opposite side due to downstream traffic spillback, it often leads to the obstruction of vehicle and pedestrian movement on conflicting approaches. The effect of "blocking the box" can propagate to nearby intersections and, in extreme cases, lead to gridlock. Along with negative traffic impacts such as capacity reductions and increased travel times, blocking the box creates potentially unsafe vehicle and pedestrian movement. A "Do Not-Block-The-Box" (DBTB) treatment seeks to reduce the likelihood of drivers entering an intersection when there is insufficient space to exit, and thus reduce blocking occurrences. DBTB treatments are typically low cost, representing a traffic management alternative accessible to most transportation agencies. This report presents a study that explored the performance of DBTB treatments installed in the greater Atlanta area.
For the studied sites, the likelihood, or propensity, of a vehicle to block was measured both before and after the DBTB treatment installation. Several blocking behavior characteristics were seen throughout the analysis. First, the change in propensity with the installation of the DBTB treatment was inconsistent, witnessing both increasing and decreasing blocking rates. However, regardless of an increase or decrease in blocking rate, the aggregated observed propensities at the studied intersections were consistently high in both the before and after treatment conditions. The lowest observed aggregate propensity
vii

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
to block was 55%, with all other time periods above 60%, and with half of the observed periods having a propensity to block of 70%. In addition, there was significant variability in day-to-day blocking opportunities. That is, at the same intersection, within the same week, peak periods where opportunities to block were rare or non-existent were observed, as well as peak periods with significant blocking opportunities. Given these findings it may not be concluded that the installed treatments reliably impacted blocking behavior. In addition, where sites did show improvements, the blocking rate remained high, often well in excess of 50%.
There are potential biases in this study. The first is that the sites were not randomly selected, but instead were identified as high blocking sites by system managers. These sites may represent the worst-case scenarios and those most difficult to address. In addition, the after data collection at several sites occurred during the I-85 bridge closure. It is not known if the potential rerouting or other driver responses to that incident may have influenced the observed driver behavior. Finally, several sites utilized police officer control during the highest demand periods. If the police were not present, these time periods may have experienced different blocking behavior than the time periods included in the evaluation. However, even given these potential biases, there still remains a failure of the DBTB treatment to address blocking under these conditions.
While not measured at these sites, it is important to highlight that a reduction in blocking may result in significant operational improvements. To explore potential performance impacts of blocking, a microscopic simulation model of blocking behavior was developed. The simulation reflects the propensity of a vehicle entering the intersection box when a blocking opportunity exists and the resulting blocking of traffic, should a
viii

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
vehicle "block the box." This model allowed for the exploration of the relationship of blocking behavior to vehicle delay and intersection capacity. It showed that the impact of blocking can be significant, potentially resulting in gridlock. However, the simulation also was able to demonstrate that DBTB treatments can significantly improve traffic flow even without achieving zero blocking. From this it may be postulated that many of the intersections that were included in this study could significantly benefit if blocking could be reduced. While the treatments in the current field study did not demonstrate the blocking rate reductions necessary for meaningful operational benefits, the simulation study does highlight the importance for continuing to seek a solution to the DBTB challenge.
While the current treatment as a standalone measure did not meaningfully impact blocking behavior, there is significant value in continuing to seek reductions in blocking behavior. Thus, based on the study findings and field observations during the data collection, several recommendations are offered.
1) Signal timing to reduce blocking opportunities. The first strategy to address blocking should be, where possible, the elimination of the potential for blocking utilizing congested period signal timing that reduces blocking opportunities. Blocking opportunities occur where the flow of vehicles into an intersection exceeds the intersection capacity, often reflected as spillback into upstream intersections. Where practical, upstream signal timing should be set to limit downstream vehicle arrivals to that of the downstream intersection processing capacity. While this may result in lower upstream performance, avoiding the gridlock within the network resulting from spillback and blocking should be prioritized. The development of such signal timings will typically require the use
ix

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
of advanced simulation tools applied at the corridor level over time periods greater than the typical peak-hour analysis. While this represents a significant investment in timing plan development, the potential benefits are substantial. 2) Reduction or elimination of free-flow turn movements during congested periods. One key observation at several sites is related to the impact of free-flow turn movements on intersection operations. Under high demand conditions a freeflow movement could continuously "fill-in" available capacity on the departure lanes of an intersection approach. This would result in vehicles from a controlled movement utilizing the same departure lanes continually being unable to proceed when they receive a green indication. It is reasonable to hypothesize that this increases the driver frustration and aggressiveness from the controlled movement, resulting in additional blocking, as this was seen as their only opportunity to proceed through the intersection. 3) Limit candidate intersections. While blocking the box occurs when a vehicle stops within the box, in a practical sense many of the observed "blocks" had minimal or no observed impacts on intersection capacity. This could occur for two reasons. The first is that the size of the intersection allowed conflicting vehicles to easily maneuver around the blocking vehicle(s). The second is that the blocking and blocked vehicles used the same intersection departure lanes. In such an instance, while blocking occurs there is no intersection capacity impact. In both situations, as drivers see minimal to no benefit of the treatment this may increase the likelihood of disregarding the treatment.
x

Operational Evaluation of Do Not Block the Box Campaigns in Georgia 4) Public education. A public education program on the benefits of not blocking the
box may help to decrease the propensity to block and reinforce the need to follow DBTB treatments. 5) Enforcement. Along with public education, there is likely a need for enforcement. For the given study, none of the intersection DBTB treatments were enforced through citations to drivers that blocked the box. The effectiveness of enforcement in improving the DBTB treatment performance should be explored. This should include different enforcement program durations (i.e., intermittent vs. continual), warnings vs. citations, automated vs. manual citations, etc. 6) Additional Treatment. Additional treatments should be developed and tested (e.g., flashing DBTB signs that are indicated only when vehicles are detected stopped in intersection departure lanes). Finally, the most probable means to successfully address intersection blocking is the development of a DBTB overarching program. Such a program would combine the above recommendations into a comprehensive strategy, developing training directed at identifying factors contributing to blocking at specific intersections, developing a signal timing plan guidance addressing blocking, designing DBTB signals and striping, and implementing widespread education, etc.
xi

: INTRODUCTION
When a vehicle enters an intersection that has insufficient space to exit on the opposite side due to downstream traffic spillback, it often leads to obstruction of vehicle and pedestrian movement on conflicting approaches. The effect of "blocking the box" can propagate to nearby intersections and, in extreme cases, lead to gridlock. Along with a negative impact on traffic, such as capacity loss and increased travel times, it also increases the potential for unsafe vehicle and pedestrian movement (1, 2).
There are several measures that are used to help prevent gridlock, including retiming signals, promoting other modes of transportation, increasing capacity of the intersection by redesigning the geometry, and installing box junctions. Box junctions, also known as "Do Not Block the Box" (DBTB) campaigns, have proven to be an economical traffic management alternative in several countries around the world, including the United States. The primary advantage of a successful DBTB campaign is that the intersection itself remains clear, even if traffic is spilling back from a downstream intersection. By preventing this downstream spillback from blocking movements from other conflicting directions, those movements with unconstrained receiving lanes should not experience any unnecessary congestion and thus avoid gridlock and the associated negative economic and traffic operations impacts.
1

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
1.1 Background
In 1964, the first box junction was installed in London and was generally seen as a successful traffic management measure (3). This idea soon spread to different countries having ties to the United Kingdom, including Ireland, Australia, and New Zealand. Later, other European and Asian countries also adopted this traffic control method. In the U.S., the first box junction was applied by the New York City Department of Transportation in 1971, where it was shown to improve traffic operations. By the 1980s, DBTB had become popular in the City of New York (4). Subsequently, the District of Columbia and the Cities of Boston and Miami aggressively adopted DBTB campaigns (3, 5, 6). The success of the DBTB campaigns in Miami and Boston can be attributed, at least in part, to extensive public/private partnerships. Recently (April 2015), a new DBTB campaign was adopted by the City of Austin, Texas (7).
These DBTB campaigns endeavor to control traffic operations by placing several signs before the intersection and at the intersection, and painting a "boundary box" on the pavement within the intersection, consistent with the standards of the MUTCD (8). Under the DBTB rules, if a vehicle is inside the box when the movement on the opposing approach has the right-of-way, it is under violation and is potentially subject to a traffic ticket, thereby creating an incentive to keep the intersection clear and reduce blocking of the opposing movements. The operational benefits of this approach are dependent upon several factors, including the location and size of the intersection, and the right-turn traffic movement policy.
2

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
1.2 Project Objectives
The overarching objective of this research is to provide the Georgia Department of Transportation (GDOT) and the Perimeter Community Improvement District (PCID) with an evaluation of the operational performance impacts of implementing a DBTB campaign at selected signalized intersections. The study also determines how variations in conditions among these intersections impact the effectiveness of DBTB. Collectively, these results are expected to aid decision-making regarding future DBTB implementations in Georgia.
The key sub-objectives of the project may be summarized as follows: Provide a review of the existing studies on "Do Not Block the Box" or "box
junction" enforcement laws, and operational performance analysis Develop a data analysis plan to identify and quantify the factors that influence the
effectiveness of DBTB at the selected intersections Extract and post-process the relevant data consistent with the above plan Perform a beforeafter analysis of the selected intersections undergoing DBTB
implementation to evaluate differences in operational performance Prepare a final report and make recommendations
The remainder of this report provides additional DBTB background, the data collection and analysis methodology (Chapter 2), findings based on field data (Chapter 3), and conclusions and recommendations (Chapter 4). A literature review chapter is not included in the report to avoid repetition of previous documentation. This project represents a continuation of GDOT project RP 13-16: Enhanced Role of Activity Center Transportation Organizations in Regional Mobility. A review of previous DBTB efforts is
3

Operational Evaluation of Do Not Block the Box Campaigns in Georgia found in that document. In addition, a detailed review of DBTB is available in the TRB paper, "Analysis of Vehicle Blocking Behavior on Intersection Performance," in Appendix A.
4

: DATA EXTRACTION AND ANALYSIS METHODOLOGY
2.1 Introduction
A vehicle entering an intersection without sufficient space to exit on the opposite side during that vehicle's phase is referred to as "blocking the box." Blocking often results in obstruction of pedestrians and other vehicles. If unchecked, intersection blocking may escalate to gridlock. For this study (with the exception of one intersection to be discussed in Chapter 3), a blocking event is considered to occur when a vehicle enters the intersection box, is not able to exit during its given phase, and obstructs conflicting phases with the right-of-way (green indication). Blocking reduces the effective green time and capacity of the obstructed movement. In this document, an approach lane that is blocked is referred to as the blocked lane and the approach lane from which the blocking vehicle enters the intersection is referred to as the block-source lane.
To study the operational impact of "Do Not Block the Box" treatments, intersection video recordings were taken during peak and adjacent off-peak demand conditions, before and after implementation of DBTB signing and marking. Data were extracted from the videos to facilitate the DBTB data analysis process. DBTB field data analysis primarily included measuring the propensity to block, i.e. the likelihood of an individual to enter an intersection when insufficient space exists to exit. Field data were used also to calibrate a VISSIM model, in which the impact on traffic characteristics, such as delay and capacity of vehicles' propensity to block, was studied in the simulation environment. This chapter discusses the data collection process, and the propensity-to-block estimation methodology.
5

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Chapter 3 then presents the findings based on the field data collection and Chapter 4 presents the VISSIM findings.
This chapter is divided further into three sections. Section 2.2: Video Data Collection Process discusses the data collection equipment, video recordings parameters, and data collection challenges. Section 2.3: Video Data Extraction Software discusses the software developed to assist in the video data extraction process, which was named the "Georgia Tech Multi Video Player" (GT-MVP). Finally, Section 2.4: Methodology to Estimate Propensity to Block presents a detailed description of the steps involved to estimate the propensity to block at an intersection.
2.2 Video Data Collection Process
In this section, the videos recorded at intersections prior to DBTB installation are referred to as before-DBTB intersection videos, and the videos recorded at intersections after DBTB installation are referred to as after-DBTB intersection videos. Not all before-DBTB intersections are included in the final after-DBTB intersection data set.
To study the impact of DBTB campaigns on blocking behavior, 712 hours of traffic videos for each of 410 days were recorded at the study intersections. Chapter 3 provides detailed lists of the intersections and the number of hours of video data collected and processed for before- and after-DBTB implementation at the study intersections. To view traffic signal indications and vehicles on the corresponding approach lanes, the video recording setup sought to capture all intersection approaches and all signal indications. To achieve this, multiple cameras (i.e., two to four) were deployed at each intersection. For five before-DBTB intersections, video data recording was outsourced to a local data collection vendor. For these intersections, the vendor recorded traffic data using four
6

Operational Evaluation of Do Not Block the Box Campaigns in Georgia cameras, each focusing on one approach/leg of the intersection. For all the remaining before- and after-DBTB intersection recordings, several customized, easily deployable, portable video data recording platforms were constructed. Each unit consisted of: 1) one GoPro HERO4 camera, 2) one 30-foot telescoping mounting pole, 3) one transparent enclosure, 4) a pair of fan units, 5) batteries, and 6) a pole support stand. The GoPro HERO4 camera, batteries, and pair of fans were placed in the transparent enclosure. A rainproof forced-air ventilation system was fabricated for the enclosure to prevent camera overheating. A unit can typically record up to 12 continuous hours of video depending on the size of the storage card. Figure 1 shows a video data recording unit deployed at the 10th Street NW and Williams Street NW intersection.
Figure 1: Video Data Recording Equipment at 10th Street NW and Williams Street NW Intersection. 7

Operational Evaluation of Do Not Block the Box Campaigns in Georgia The research team developed the upgraded video platform to enhance the quality of
the video recordings, as reliable identification of the signal indications in the initial video proved challenging with standard definition videos in the commercially available solutions. In addition, the developed platform had a wider field of view, reducing the number of cameras needed at each intersection, which aided in reducing the data extraction effort. Figure 2 shows a comparison of image quality between video recordings captured by the standard definition camera units used by the vendor and the Georgia Tech Portable Video Data Recording Platform.
Based on file size limitations, videos recorded by the data collection vendor and by the Georgia Tech Portable Video Recording Platform were segmented into 30-minute and 15-minute video files, respectively. At the start of each data collection day using the Georgia Tech Portable Video Recording Platform, the SD card in each camera was replaced with an empty card and the previous day's batteries were replaced with charged batteries.
8

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
(a)
(b) Figure 2: Enhancement in Video Quality at Intersection of Peachtree Dunwoody Rd.
at Johnson Ferry Rd., Comparing: (a) Standard Definition Vendor Unit, and (b) Georgia Tech's Portable Video Recording Platform. 9

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
2.3 Video Data Extraction Software
This section discusses the software used to extract data from the video recordings. To estimate propensity to block, two sets of data elements were extracted: 1) blocking event data, and 2) Block-Source Lane vehicle categorization data. Blocking event data extraction included documenting blocking event details, such as start of blocking event timestamp, end of blocking event timestamp, Block-Source Lane, Blocked Lane, etc. Block-Source Lane identification reviewed all traffic on the Block-Source Lane to identify both the vehicles that blocked and the vehicles that chose not to block, allowing for the calculation of propensity to block. The following paragraphs provide a detailed description of the data extraction steps.
Initially, blocking event data were extracted by reviewing the video and manually entering timestamps and other data into customized Microsoft Excel spreadsheets (Figure 3). A Java-based computer software program, VideoAnalyzer, developed in 2014 by the School of Civil and Environmental Engineering at Georgia Tech, was used to retrieve vehicle timestamps (Figure 4). A detailed discussion of VideoAnalyzer and its application may be found in the thesis, "Traffic Management Alternatives for Business Improvement Districts" (5). However, the researchers in this study found the VideoAnalyzer software to contain a number of inefficiencies that hindered the ability to complete the project in a timely manner. To improve the efficiency and accuracy of the data extraction process, an additional software tool, the Georgia Tech Multi Video Player, was developed. Table 1 lists the limitations of VideoAnalyzer and the upgrades found in GT-MVP. A detailed discussion of GT-MVP may be found in Saroj et al., "Video Tool for Manually Extracting Complex Traffic Data," provided in Appendix C (9).
10

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure 3: Snapshot of Customized Blocking Event Extraction Sheet.
Figure 4: Example Timestamp Extraction Using VideoAnalyzer for Vehicle Entering the Intersection Timestamp Taken When Vehicle in Blue Box Crosses the Crosswalk. 11

Operational Evaluation of Do Not Block the Box Campaigns in Georgia

Table 1: Comparison of VideoAnalyzer and GT-MVP Upgrades.

Limitations of VideoAnalyzer

New Video Software Upgrades GTMVP

Unable to play multiple videos in sync

Enables playing two videos in sync

No skip or forward button

Enables customized skipping forward/backward

First timestamp taken from PC internal

clock, subsequent timestamps represent Timestamp read from video

difference from video start.

The GUI does not contain: entry widgets to enter specific details ability to clear/undo an error

Provides a customized GUI to extract data
Propensity extraction
Blocking event extraction

Requires manual entry of data into Writes and reads to csv files

spreadsheet

automatically

GT-MVP contains two video player windows and one data extraction window (Figure 5). Each video player window contains a time slider (allowing the user to jump to a time in the video) and basic control buttons: Play, Pause, Play (0.5X), Play (2X), and a customizable forward/backward button. The customizable forward/backward button allows the user to choose a custom value (in seconds) to skip back or forth in the video.

12

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure 5: Screen Capture of GT-MVP Software Interface. GT-MVP's data extraction window also contains video player control buttons: Play/Pause, Play(1X), Play(2X), customized Forward Skip and Backward Skip, and Get Timestamp. These buttons control both videos simultaneously, enabling a user to play two videos in a synchronized manner. While the VideoAnalyzer's timestamp button extracts the PC clock time, GT-MVP's Get Timestamp button retrieves the "time into video" timestamp (i.e., the time when the video recording started, encoded in the filename, is added to the "time into video" to obtain the correct time of day). The graphical user interface (GUI) of the data extraction window contains an editable cell matrix to record data. GT-MVP populates these cells automatically when the user clicks the Get Timestamp button on the data extraction interface, but the cells are editable to allow for correction of errors. The cell matrix section is scrollable, similar to an Excel spreadsheet. Additionally, the Save to CSV and Load CSV buttons allow a user to save the extracted data in cell matrix to a csv file and load the data from a csv file to the cell matrix,
13

Operational Evaluation of Do Not Block the Box Campaigns in Georgia respectively. Two different data extraction windows were designed for GT-MVP. One was customized for extracting blocking event data and the other for extracting the Block-Source Lane categorization data. (Figure 6). This software is shared for community use at http://transportation.ce.gatech.edu/publications.
14

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
(a)
(b) Figure 6: Snapshots of (a) GT-MVP's Blocking Event Data Extraction Window (9), and (b) GT-MVP's Block-Source Lane Categorization Data Extraction Window (9).
15

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
2.4 Methodology to Estimate Propensity to Block
This section discusses the methodology to estimate propensity to block. This methodology can be divided into four steps, the first two steps extracting the blocking event data and the final two steps extracting the Block-Source Lane categorization data.
In step 1, Initial Review, intersection video recordings are reviewed for the presence of blocking events. Video recordings containing blocking are flagged for further processing.
In step 2, Blocking Event Extraction, blocking event details, such as blocking event start time, blocking event stop time, Block-Source Lane, Blocked Lane, and if the blocked lanes were fully or partially blocked, are extracted from the video recordings flagged in step 1.
In step 3, a detailed Block-Source Lane categorization data collection plan is determined for each intersection included in step 2. As collection of all potential data exceeded project resources, the research team selected a subset of days, times, and lanes for the propensity to block data extraction (i.e., Block-Source-Lane categorization) based on the extracted blocking event details from step 2. For each intersection, for the days and lanes selected, the full peak period would be included in the Block-Source Lane categorization data extraction to ensure capture of instances when a vehicle chose not to block when a blocking opportunity existed.
In step 4, a vehicle on the Block-Source Lane was categorized as a) a blocking vehicle, if it entered the intersection resulting in a block; or b) a non-blocking vehicle, if the vehicle had the opportunity to block and did not enter the intersection, but instead waited for the next cycle to avoid blocking. The ratio of blocking vehicles to the number
16

Operational Evaluation of Do Not Block the Box Campaigns in Georgia of vehicles with an opportunity to block (i.e., number of blocking vehicles + number of non-blocking vehicles) was estimated. This ratio is the estimated propensity to block for the corresponding Block-Source Lane for the study intersection. Figure 7 shows an overview of this four-step process. A detailed discussion of each step is presented in the following subsections.
Figure 7: Flow Chart of Methodology Used to Estimate Propensity to Block. 2.4.1 Step 1. Initial Review
As a first step, a review of the video recordings was undertaken to check for the presence of blocking events. Video recordings of the peak period for each intersection were reviewed first. If no blocking (or significant congestion) was seen on an intersection's peak period video recordings, the remainder of the video recordings for that day were not reviewed. If blocking was noted during the peak period, the video recording review was expanded beyond the peak until the video recordings did not contain blocking or all video recordings had been checked. The videos that were found to have blocking events were
17

Operational Evaluation of Do Not Block the Box Campaigns in Georgia flagged, as shown in Figure 8. The data for the step 2 analysis were extracted from these videos.
Figure 8: Snapshot of the Spreadsheet Used to Record Flagged Videos in the Initial Review Process.
2.4.2 Step 2. Blocking Event Data Extraction In this step, the details of blocking events observed in step 1 were extracted using the
GT-MVP software. As discussed earlier, for this study, a blocking event occurs when a vehicle enters and remains in the intersection through a phase change, thus blocking a green indication on an opposing movement. In this analysis, presence of a vehicle on the blocked approach is not necessary for the block to be recorded. The blocking events were observed in videos from multiple views of the intersection to collect the necessary data. Figure 9 shows a snapshot of a blocking event extraction sheet.
18

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure 9: Snapshot of Blocking Event Extraction Sheet for the Intersection of Clairmont Rd. at I-85 SB, near Sam's Club, 04-17-2017 18:11
(Video 1: GP113441_20170417_181134, Video 2: GP083445_20170417_182317). The following details for each blocking event were extracted:
Block Event Start Time: The timestamp of the start of the blocking event. The timestamp from the start of the video was recorded, with the video start as time t = 0. This time was then converted to the time of day by adding the start time of the video recording.
Block Event Stop Time: The timestamp of the end of the blocking event. As with block event start time, the timestamp from the end of the video was recorded, with the video start as time t = 0. This time was then converted to the time-of-day by adding the start time of the video recording. 19

Operational Evaluation of Do Not Block the Box Campaigns in Georgia Block-Source Lane Number and Blocked Lane Number: The vehicle movements involved in the blocking event, i.e., the movement originating the blocking vehicle (Block-Source Lane) and the movement being blocked (Blocked Lane), were identified for each blocking event. For efficient data recording, all lane movements for each intersection were coded according to a three-digit number. An example of the coding for Peachtree Dunwoody Rd. at Johnson Ferry Rd. is shown in Figure 10. The coding schemes for each intersection analyzed can be found in Appendix B.
Figure 10: Lane Movement Numbers Assigned at Peachtree Dunwoody Rd. at Johnson Ferry Rd. Vehicle Obstruction: "Yes" or "No" under the "Vehicle Obstruction" column was used to indicate if a blocking vehicle impacted the regular movement of vehicles for each blocking event. Examples of "Yes" and "No" blocking events are described below:
20

Operational Evaluation of Do Not Block the Box Campaigns in Georgia o Yes: Indicates a blocking event with an obstruction of the conflicting
vehicle's travel path. Figure 11 is a screen capture of a blocking event at the intersection of Peachtree Dunwoody Rd. and Abernathy Rd. that started at 16 hr. 02 min. 52 sec. The red arrow indicates the movement of vehicles from the Block-Source Lane (Northbound (NB) left), and the blue arrow indicates the movement of the vehicles in the Blocked Lane (Eastbound (EB) left). The red boxes indicate the blocking vehicles and the blue box indicates the first blocked vehicle associated with the blocking event. During the video observation, the blocking vehicles clearly obstruct the right-of-way of the blocked vehicles.
Figure 11: Vehicles in Red Blocked the Vehicles Marked in Blue (Peachtree Dunwoody Rd. at Abernathy Rd. NE, 10/28/2015, 17:02).
21

Operational Evaluation of Do Not Block the Box Campaigns in Georgia o No: Indicates a blocking event with no obstruction of the conflicting vehicle's travel path. Figure 12 provides an example of a blocking event captured in the same video as above, at the intersection of Peachtree Dunwoody Rd. and Johnson Ferry Rd. In this image the white pickup (in the red rectangle) is present in the intersection while the green indication for the Southbound (SB) left and SB through movements are active (indications are highlighted in the yellow box). However, neither the SB left nor through movement is obstructed by this blocking event. This is primarily a result of the large size of the intersection "box."
Figure 12: Example Blocking Event that Does Not Obstruct Traffic Flow (Peachtree Dunwoody Rd. at Johnson Ferry Rd., 10/26/2015, 16:00 1_20151026_160000_0001n0_cam3.avi).
Presence of Demand: The presence of demand on the Blocked Lane was also captured under the Presence of Demand column. "Yes" indicates that demand is present and "No" indicates that no demand is present. For example, in both examples above (Figure 11 and Figure 12) the SB left turn (Blocked Lane) has
22

Operational Evaluation of Do Not Block the Box Campaigns in Georgia demand during the blocking event. This is indicated with a "yes" under the Presence of Demand column for the blocking event extraction. Figure 13 below shows a blocking event captured at the intersection of Ashford Dunwoody Rd. NE at Ravinia Dr. NE. In this example, the white vehicle (identified by the red rectangle) is blocking the crossing approach that has a green indication (signal heads marked with a smaller red rectangle). However, there is no vehicle present on the blocked lane (yellow arrow points to the blocked lane), indicating an absence of demand.
Figure 13: Blocking Event with No Presence of Demand (Ashford Dunwoody Rd. NE at Ravinia Dr. NE, 10/29/2015, 16:39). Capacity Impact (Full/Partial): Blocking events that affected capacity were further categorized as a full block or partial block.
23

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
o Full Block: For a full blocking event, the blocked vehicles are unable to complete their movement.
o Partial Block: For a partial blocking event, the blocked vehicles are hindered but are able to successfully complete the desired movement.
Comments: Under the comments section, additional details about the blocking event are documented.
2.4.3 Step 3. Day, Time, and Lane Selection In step 3, first, the blocking event details extracted for before- and after-DBTB
treatment at an intersection were inspected to note a 3-hour time duration that captures most of the blocking events on all the days. Next, the days with the maximum number of blocking events with obstruction of right-of-way (i.e., capacity impact) for the selected 3hour time duration were identified. Lastly, Block-Source Lanes corresponding to the blocking events observed during the selected time period on the selected days were identified. These time periods were further analyzed in step 4 to allow for a determination of propensity to block on the identified Block-Source Lanes. All time periods including blocks were not included in the propensity analysis due to time limitations in data extraction.
2.4.4 Step 4. Block-Source Lane Vehicle Categorization In step 4, the blocking behavior of vehicles on the identified Block-Source Lanes for
the selected days and time durations was inspected. A vehicle's choice to enter or to not enter the intersection, when presented with an opportunity to block (i.e., no space downstream to exit the intersection), was recorded. Vehicles on the Block-Source Lane
24

Operational Evaluation of Do Not Block the Box Campaigns in Georgia were annotated by recording the timestamp when the vehicle crossed the stop bar to enter the intersection. A vehicle was designated to be a blocking vehicle if it entered the intersection and created a block. A vehicle was designated to be a non-blocking vehicle if it chose not to enter the intersection while having a green indication. Thus, the designation of non-blocking was based on the judgement of the vehicle driver, not the data analyst. Vehicles that entered the intersection and did not create a block were not included in the blocking or non-blocking designations. Figure 14 shows a snapshot of GT-MVP's propensity extraction module being used for the Block-Source Lane vehicle categorization for lane number 623 at the intersection of Peachtree Dunwoody Rd. and Johnson Ferry Rd. Figure 14 also includes a snapshot of the generated Excel spreadsheet. The vehicle categorization into 1) blocking vehicle and 2) non-blocking vehicle can be seen under the Vehicle Type column (marked in the blue rectangle).
Figure 14: Snapshot: Propensity Extraction Interface of GT-MVP (Peachtree Dunwoody Rd. at Johnson Ferry Rd., 04-27-2016, 16:22 PM
GP062213/GP070993) with Extracted Propensity Data Spreadsheet.
25

Operational Evaluation of Do Not Block the Box Campaigns in Georgia Finally, propensity to block for the intersection was estimated by determining the
ratio of number of blocking vehicles to the total number of vehicles with an opportunity to block.
# #
26

: DATA AND FINDINGS
This study included 17 intersections in the before phase of the DBTB data collection effort. Of these, 9 intersections were included in the after study. The selection of the original 17 intersections was based on discussions with GDOT, GDOT contractors, and the PCID, identifying locations with a high likelihood of blocking. After the selection of the 17 study intersections, GDOT personnel designed the DBTB treatments for each intersection and programmed the treatment installations. Eleven of the 17 intersections were excluded from the after study due to: 1) a lack of observed blocking in the before data (six intersections), 2) the treatment not being installed in time to be included in the study (two intersections), or 3) a reduction in data extraction efforts due to project resource limitations (three intersections). Intersections included in the after study were observed to have at least several instances of blocking on at least one movement in the before data extraction. The following sections review the intersections selected, the data collection, and the findings.
3.1 Intersection Selection
The intersections in the study included a set of 5 intersections within the Perimeter Community Improvement District and 12 additional intersections on state routes throughout the Atlanta region (referred to as GDOT intersections in this report).
3.1.1 PCID Before Data at the five PCID intersections were collected during October 2015. Intersection operations were video recorded Monday to Thursday (or Friday) for a single week, with
27

Operational Evaluation of Do Not Block the Box Campaigns in Georgia

two weeks' data collected at several intersections. For each intersection, video recordings were taken from approximately 3 PM to 9 PM, covering the PM peak and several hours before and after the peak. An external vendor completed these video recordings. Table 2 provides a list of the PCID intersections, dates, and days of video recording for the before data collection.

Table 2: PCID Before DBTB Installation: Intersections and Video Recording Dates and Days.

Intersection Name

Dates

Days

1. Parkside PI at Perimeter Summit Pkwy.

10/19/201510/23/2015 10/26/201510/29/2015

MonFri MonThu

2. Ashford Dunwoody Rd. NE at Ravinia Dr. NE

10/26/201510/29/2015

MonThu

3. Peachtree Dunwoody Rd. at 10/19/201510/23/2015

Lake Hearn Dr. NE

10/26/201510/29/2015

MonFri MonThu

4. Peachtree Dunwoody Rd. at 10/26/201510/29/2015 Abernathy Rd. NE

MonThu

5. Peachtree Dunwoody Rd. at 10/26/201510/29/2015 Johnson Ferry Rd.

MonThu

3.1.2 PCID After The after study intersections were selected based on the nature of blocking events observed in the before study intersection video recordings. Of the initial five intersections, four intersections were observed to experience blocking with an impact on the capacity of the intersection. Only the Parkside Pl at Perimeter Summit Pkwy intersection had limited blocking events and was not carried forward in the after study, as the blocking observed did not impact intersection capacity. Similar to the before video recordings, the after video recordings covered the weekdays Monday to Thursday, from approximately 3 PM to 9 PM.

28

Operational Evaluation of Do Not Block the Box Campaigns in Georgia

The after video recordings were completed by the Georgia Tech research team using the equipment discussed in Chapter 2. Table 3 shows a list of the PCID intersections, dates, and days of data collection for the after data collection.

Table 3: PCID After DBTB Installation: Intersections and Video Recording Dates and Days.

Intersection Name

Dates

Days

1. Ashford Dunwoody Rd. NE at Ravinia Dr. NE

04/18/201604/21/2016 MonThu

2. Peachtree Dunwoody Rd.at Lake Hearn Dr. NE

04/25/201604/29/2016 05/06/2016

MonFri Fri

3. Peachtree Dunwoody Rd. at Abernathy Rd. NE

04/18/201604/21/2016

MonThu

4. Peachtree Dunwoody Rd. at Johnson Ferry Rd.

04/25/201604/29/2016

MonThu

Figure 15 provides an example aerial view of Ashford Dunwoody Rd. at Ravinia Dr. NE prior to DBTB implementation. Figure 16 and Figure 17 show example image captures from the after video recordings for this intersection. An aerial image, an image from each video recording angle with and without treatment, and a sketch showing the lane numbering (as discussed in Chapter 2) for each intersection are provided in Appendix B.

29

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure 15: Ashford Dunwoody Rd. NE at Ravinia Dr. NE Source: Google Street View.
Figure 16: Ashford Dunwoody Rd. NE at Ravinia Dr. NE, Video Screen Capture, Location 1, 04/21/2016. 30

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure 17: Ashford Dunwoody Rd. NE at Ravinia Dr. NE, Video Screen Capture, Location 2, 04/21/2016.
3.1.3 GDOT Before Along with the PCID locations, GDOT selected 12 additional intersections for DBTB treatment. These intersections were video recorded for before DBTB installation in March and April of 2016. Video recording was completed by Georgia Tech staff, recording each intersection for at least four out of five weekdays. Table 4 lists the 12 intersections and video recording dates and days.
31

Operational Evaluation of Do Not Block the Box Campaigns in Georgia

Table 4: Additional GDOT Before Intersections and Video Recording Dates and Days.

Intersection Name

Dates

Days

1. Peachtree Rd. NE at Lenox Rd. 03/15/201603/18/2016 NE

TueFri

2. Peachtree Rd. NE at Highland Dr. NE

03/15/201603/18/2016

TueFri

3. Peachtree Rd. NE at Stratford Rd. NE

03/15/201603/18/2016

TueFri

4. Ponce De Leon Ave. NE at City 03/15/201603/18/2016 Hall East

TueFri

5. S. Cobb Dr. at Pearl St.

03/21/201603/25/2016 MonFri

6. S. Cobb Dr. at Walker St.

03/21/201603/25/2016 MonFri

7. W. Paces Ferry Rd. NW at I-75 SB On/Off Ramps
8. Mount Paran Rd. NW at I-75 NB Off Ramp
9. Williams St. NW at 10th St. NW

03/28/201604/01/2016 03/28/201603/31/2016 04/05/201604/08/2016

MonFri MonThu TueFri

10. South Cobb Dr. at I-285 SB On/Off-Ramps
11. Clairmont Rd. at I-85 SB, near Sam's Club
12. 14th St. NW at Hemphill Ave. NW

04/04/201604/08/2016 04/11/201604/15/2016 04/11/201604/15/2016

MonFri MonFri MonFri

3.1.4 GDOT After As with the PCID intersections, the after study intersections were primarily selected based on the nature of blocking events observed in the before study intersection video recordings. These 12 intersections were observed and categorized in "High, Medium, and Low" levels

32

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
of blocking in terms of frequency and duration of the blocking event. Due to limitations in data extraction project resources, intersections with a "Low" level of blocking were not included in the after analysis. Of the initial 12 intersections, 7 were found to have sufficient blocking in the before period to warrant inclusion in the after DBTB analysis. In this instance, insufficient blocking was defined as either few blocking instances during the peak periods or blocking with no significant capacity impact (i.e., a blocking vehicle may encroach into the intersection box but does not hinder any conflicting vehicles). Table 5 lists those intersections not included in the after analysis. Of the remaining seven intersections where blocking was observed, the intersections of South Cobb Dr. at Walker St. and South Cobb Dr. at I-285 SB On/Off Ramps were not included, as the DBTB treatment had not yet been installed at the time of the after data video recording. Similar to the before video recordings, the after video recordings occurred during weekdays, starting approximately from 24 PM until 10 PM. The after video recordings were completed by the Georgia Tech research team. A list of the GDOT intersections, dates, and days of data collection are given in Table 6 for the after data collection.
Table 5: Intersections Dropped from the Study. Intersection Name Peachtree Rd. NE at Lenox Rd. NE Peachtree Rd. NE at Highland Dr. NE Ponce De Leon Ave. NE at City Hall East S. Cobb Dr. at Pearl St. Mount Paran Rd. NW & I-75 NB Off Ramp
33

Operational Evaluation of Do Not Block the Box Campaigns in Georgia

Table 6: Additional GDOT After Intersections and Video Recording Dates and Days.

Intersection Name

Dates

Days

1. Peachtree Rd. NE at Stratford Rd. NE
2. W. Paces Ferry Rd. NW at I-75 SB On/Off-Ramps
3. Williams St. NW at 10th St. NW
4. Clairmont Rd. at I-85 SB, near Sam's Club
5. 14th St. NW at Hemphill Ave. NW

03/29/201703/31/2017, 04/19/201704/20/2017, 04/25/2017
04/12/201704/13/2017, 04/17/201704/18/2017, 4/27/2017
04/25/201705/27/2017, 05/01/2017, 05/03/2017, 05/08/2017
03/29/201703/31/2017, 04/12/201704/13/2017, 04/17/201704/18/2017, 04/21/2017
04/20/2017, 05/01/2017, 05/03/2017, 05/08/201705/09/2017

TueFri MonThu MonThu MonFri
MonThu

Of the five intersections in Table 6, data extraction and analysis are included in the findings for the intersections of West Paces Ferry Rd. NW at I-75 SB On/Off Ramps and Clairmont Rd. at I-85 SB, near Sam's Club. The remaining three intersections are not included in the final results due to the project's budget limitations. However, a visual review of the intersection videos indicates findings likely similar to the other seven intersections fully reduced for this study. An aerial image, an image from each video recording angle with (where applicable) and without treatment, and a sketch showing the lane numbering for each intersection may be found in Appendix B.

3.2 Intersection Results
As discussed in section 2.4, propensity analysis was undertaken at intersections with at least a minimum threshold of blocking events in the before data. For selected intersections,

34

Operational Evaluation of Do Not Block the Box Campaigns in Georgia on the days of interest, data were extracted for a 3-hour period, unless otherwise noted. The 3-hour range was selected to encompass any observed blocking. The same 3-hour window was utilized in the before and after data collection. Extracting data over the full 3-hour block typically included video with no blocking. However, it was critical to obtaining an accurate propensity estimate to include the full 3-hour period as instances of vehicles choosing not to block may be present during these no-block present videos. The intersections and monitored lanes included in the final propensity analysis are indicated in Table 7 below. For each intersection, data were collected for 3 days in both the before and after period, with an additional fourth day at the intersections of Peachtree Dunwoody Rd. at Abernathy Rd. NE and Peachtree Dunwoody Rd. at Johnson Ferry Rd. All lanes that experienced blocking on any of the days either before or after were included in the analysis with the exception that in both Peachtree Dunwoody Rd. at Abernathy Rd. NE and Peachtree Dunwoody Rd. at Johnson Ferry Rd. a lane with a single block over the entire time period was not included in order to reduce data extraction efforts. Appendix B contains the lane configuration map and numbering key for each intersection. A discussion of each intersection's results is presented in the following sections.
35

Operational Evaluation of Do Not Block the Box Campaigns in Georgia

Table 7: Final Propensity Analysis Intersections and Lanes.

Intersection Name
1. Ashford Dunwoody Rd. NE at Ravinia Dr. NE
2. Peachtree Dunwoody Rd. at Lake Hearn Dr. NE
3. Peachtree Dunwoody Rd. at Abernathy Rd. NE
4. Peachtree Dunwoody Rd. at Johnson Ferry Rd.
5. W. Paces Ferry Rd. NW at I-75 SB On/Off-Ramps
6. Clairmont Rd. at I-85 SB, near Sam's Club

Lanes Selected 222, 223, 224, 225, 711, 712
311, 623, 511, 711, 222
311, 111, 112, 312
311, 312, 623, 624
622, 711, 712, 623, 222, 223, 511 111, 222, 223, 224, 711, 831

3.2.1 Ashford Dunwoody Rd. NE at Ravinia Dr. NE For the Ashford Dunwoody Rd. NE and Ravinia Dr. NE intersection, 4 hours of propensity data were extracted from the before video recordings, 2 hours on 10/26/2105, 1 hour on 10/27/2015, and 1 hour on 10/28/2015. Three-hour periods were not extracted for these days as police were providing traffic control during all other high-volume periods with the potential for blocking. For the after study, 9 hours of data were extracted from the video recordings over 3 days, from 4:307:30 PM each day. Table 8 provides the number of hours of data extracted for each day, total number of blocking opportunities, total number of observed blocking vehicles, total number of observed non-blocking vehicles, and blocking propensity. The analysis is provided for each day of observation and overall before and after.

36

Operational Evaluation of Do Not Block the Box Campaigns in Georgia

Table 8: Ashford Dunwoody Rd. NE at Ravinia Dr. NE.

Before
Hours of Study Total # Opportunities Total # Blocking Total # Non-Blocking Propensity

10/26/2015 (Mon*) 2
23
19
4
0.83

10/27/2015 (Tue*) 1
1
0
1
0.00

10/28/2015 (Thu*) 1
4
1
3
0.25

Total
4 28 20 8 0.71

After
Hours of Study Total # Opportunities Total # Blocking Total # Non-Blocking Propensity * Police present

04/18/2016 04/19/2016 04/21/2016

(Mon)

(Tue)

(Thu)

3

3

3

9

10

154

140

304

5

79

100

184

5

75

38

118

0.50

0.51

0.71

0.61

In the before analysis, the highest observed propensity to block was 0.83, on Monday (10/26/2015), whereas the lowest observed propensity was 0.0 on Tuesday (10/27/2015), although only a single opportunity to block was observed, thus the value lacks significance. Aggregating overall blocking and non-blocking opportunities during the before data collection, a propensity to block of 0.71 was observed. That is, 71% of drivers chose to block when presented with a blocking opportunity. In the after DBTB implementation, the highest observed propensity to block during a 3-hour period was 0.71, on Thursday (04/21/2016), whereas the lowest observed propensity to block was 0.50, on Monday (04/18/2016). The aggregate after propensity to block was 0.61, or 61%, which was 10% lower than the before period.
This intersection had several characteristics in blocking behavior that will be seen throughout the analysis. First, even though a decrease in propensity to block was observed,

37

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
the after blocking rate remained high, with a higher likelihood that a vehicle would choose to block than to not block. Also, there was significant variability in day-to-day blocking opportunities. Within the same week there would be peak periods where the opportunity to block was rare or non-existent, while on other days significant (hundreds) of blocking opportunities were observed. Traffic counts were conducted on the video data to explore a potential relationship between intersection counts and blocking. However, no meaningful relationship was found, as blocking is a function of spillback emanating from downstream intersections or system congestion. The researchers were not able to collect meaningful data on the overall system demands, as for any given day only one or two intersections were under observation. However, Appendix A, discussed in Section 3.4, presents simulated findings based on propensity to block that demonstrate the impact of system demand.
3.2.2 Peachtree Dunwoody Rd. at Lake Hearn Dr. NE For the Peachtree Dunwoody Rd. at Lake Hearn Dr. intersection, 9 hours of propensity data were extracted from the before video and 6.5 hours of propensity data were extracted from the after video. Similar to the Ashford Dunwoody Rd. NE at Ravinia Dr. intersection, the after data were limited to under 3 hours in two peak periods due to the presence of traffic control police officers at the intersection. Videos during periods when a traffic control police officer was present, was not used for data extraction. All video extraction was between the hours of 3:30 PM and 6:30 PM. Table 9 provides the number of hours of data extracted for each day, total number of blocking opportunities, total number of observed blocking vehicles, total number of observed non-blocking vehicles, and blocking
38

Operational Evaluation of Do Not Block the Box Campaigns in Georgia

propensity. The analysis is provided for each day of observation and overall before and after.

Table 9: Peachtree Dunwoody Rd. at Lake Hearn Dr. NE.

Before
Hours of Study Total # Opportunities Total # Blocking Total # Non-Blocking Propensity

10/22/2015 (Thu) 3
67
47
20
0.70

10/26/2015 (Mon) 3
40
30
10
0.75

10/29/2015 (Thu) 3
7
0
7
0

Total
9 114 77 37 0.68

After
Hours of Study Total # Opportunities Total # Blocking Total # Non-Blocking Propensity * Police present

04/26/2016 (Tue*) 1.5
2
2
0
1

04/28/2016 05/06/2016

(Thu)

(Mon*)

3

2

6.5

30

12

44

26

11

39

4

1

5

0.87

0.92

0.89

The highest observed before study propensity to block was 0.75, on Monday (10/26/2015), whereas the lowest propensity of 0.0 was observed on Thursday (10/29/2015), although the total opportunities were limited in that instance to seven opportunities. The highest observed after propensity to block was 1.0 on Tuesday (04/26/2016), again of limited significance as there were only two blocking opportunities, and the lowest was 0.87 on Thursday (04/28/2016). Aggregating overall blocking and nonblocking opportunities during the before and after data collection, propensities to block of 0.68 and 0.89 were observed. Unlike the Ashford Dunwoody Rd. NE at Ravinia Dr. NE, in this instance, the propensity to block increased. However, similar to the Ashford

39

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Dunwoody Rd. NE at Ravinia Dr. NE, both the before and after propensity rates were high, with drivers in both cases more likely to block than not block. And again, from day to day, significant variations were seen in the likelihood to block.
3.2.3 Peachtree Dunwoody Rd. at Abernathy Rd. NE For the Peachtree Dunwoody Rd. and Abernathy Rd. NE intersection, 8 hours of propensity data were extracted from the before video, and 4 hours of propensity data were extracted from the after video. Again, times for extraction were limited by the presence of traffic police. The 3-hour beforeafter study period was 4:007:00 PM. The highest observed before study propensity to block was 0.93 on Thursday (10/29/2015), whereas the lowest propensity of 0.75 was observed on Tuesday (10/27/2015), although the total opportunities were limited in that instance to four opportunities. For this intersection, the limited availability of after data due to the presence of traffic police resulted in inconclusive findings, with only 12 opportunities to block observed over the 4-day period. However, as with the previous two intersections, it is again seen in the before data that propensity rates were high, with drivers consistently more likely to block than not block.
Table 10 provides the number of hours of data extracted for each day, total number of blocking opportunities, total number of observed blocking vehicles, total number of observed non-blocking vehicles, and blocking propensity. The analysis is provided for each day of observation and overall before and after.
The highest observed before study propensity to block was 0.93 on Thursday (10/29/2015), whereas the lowest propensity of 0.75 was observed on Tuesday (10/27/2015), although the total opportunities were limited in that instance to four
40

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
opportunities. For this intersection, the limited availability of after data due to the presence of traffic police resulted in inconclusive findings, with only 12 opportunities to block observed over the 4-day period. However, as with the previous two intersections, it is again seen in the before data that propensity rates were high, with drivers consistently more likely to block than not block.

Table 10: Peachtree Dunwoody Rd. at Abernathy Rd. NE.

Before
Hours of Study Total # Opportunities Total # Blocking Total # NonBlocking Propensity

10/26/2015 (Mon*) 2.5
62

10/27/2015 (Tue*) 2
4

10/28/2015 (Wed*) 1.5
19

10/29/2015 (Thu*) 2
45

51

3

16

42

11

1

3

3

0.82

0.75

0.84

0.93

Total
8 130
112 18
0.86

After
Hours of Study Total # Opportunities Total # Blocking Total # NonBlocking Propensity

04/18/2016 (Mon*) 1
3

04/19/2016 (Tue*) 1
8

04/20/2016 (Wed*) 1
1

0

0

1

3

8

0

0

0

1.0

* Police present Data insufficient to conclude

04/21/2016 (Thu*) 1 0
0 0
NA

4 12
1 11
Insufficient Data

3.2.4 Peachtree Dunwoody Rd. at Johnson Ferry Rd. For the Peachtree Dunwoody Rd. and Johnson Ferry Rd. intersection, 10.5 hours of propensity data were extracted from the before video and 8.75 hours of propensity data

41

Operational Evaluation of Do Not Block the Box Campaigns in Georgia

were extracted from the after video. Again, times for extraction were limited by the presence of traffic police. The 3-hour beforeafter study period was 3:006:00 PM. Table 11 provides the number of hours of data extracted for each day, total number of blocking opportunities, total number of observed blocking vehicles, total number of observed nonblocking vehicles, and blocking propensity. The analysis is provided for each day of observation and overall before and after.

Table 11: Peachtree Dunwoody Rd. at Johnson Ferry Rd.

Before

Hours of Study

Total

#

Opportunities

Total # Blocking

Total # Blocking
Propensity

Non-

10/26/2015 (Mon*) 2 187
161 26
0.86

10/27/2015 (Tue) 3 1
1 0
0.00

10/28/2015 (Wed) 3 0
0 0
NA

10/29/2015 (Thu*) 2.5
31

Total
10.5 219

20

182

11

37

0.65

0.83

After Hours of Study

04/25/2016 04/26/2016 04/27/2016 04/28/2016

(Mon)

(Tue)

(Wed*)

(Thu)

1.5

2

2.25

3

8.75

Total

#

0

0

Opportunities

Total # Blocking

0

0

84

59

143

50

39

89

Total # Non-

0

0

34

20

54

Blocking

Propensity

NA

NA

0.60

0.66

0.62

* Police present

The highest observed before study propensity to block was 0.86, on Monday (10/26/2015), whereas the lowest propensity of 0.0 was observed on Tuesday (10/27/2015), although again the total opportunities were limited in this instance to 1 opportunity. The highest observed after propensity to block was 0.66 on Thursday (04/28/2016) and the

42

Operational Evaluation of Do Not Block the Box Campaigns in Georgia lowest was 0.60 on Wednesday (04/27/2016). Aggregating overall blocking and nonblocking opportunities during the before and after data collection, propensities to block of 0.83 and 0.62 were observed, a 21% decrease in propensity. However, similar to the previous intersections, both the before and after propensity rates were high, with drivers in both cases more likely to block than not block. And again, from day to day, significant variations were seen in the likelihood to block.
3.2.5 W. Paces Ferry Rd. NW at I-75 SB On/Off Ramps For the West Paces Ferry Rd. NW and I-75 SB On/Off Ramps intersection, 9 hours of propensity data were extracted from both the before and after videos. The 3-hour before after study period was 4:007:00 PM. Table 12 provides the number of hours of data extracted for each day, total number of blocking opportunities, total number of observed blocking vehicles, total number of observed non-blocking vehicles, and blocking propensity. The analysis is provided for each day of observation and overall before and after.
43

Operational Evaluation of Do Not Block the Box Campaigns in Georgia

Table 12: W. Paces Ferry Rd. NW at I-75 SB On/Off Ramps.

Before
Hours of Study Total # Opportunities Total # Blocking Total # Non-Blocking Propensity

03/29/2016 (Tue) 3
34
18
16
0.53

03/30/2016 (Wed) 3
46
26
20
0.57

04/01/2016 (Fri) 3
5
3
2
0.60

Total
9 85 47 38 0.55

After
Hours of Study Total # Opportunities Total # Blocking Total # Non-Blocking Propensity

04/17/2017 (Mon) 3
4
3
1
0.75

04/18/2017 04/27/2017

(Tue)

(Thu)

3

3

9

8

21

33

5

14

22

3

7

11

0.63

0.67

0.67

The highest observed before study propensity to block was 0.60, on Friday (04/01/2016), whereas the lowest propensity of 0.53 was observed on Tuesday (03/29/2016). The highest observed after propensity to block was 0.75 on Monday (04/17/2017) and the lowest was 0.63 on Tuesday (04/18/2017), again with a low sample size. Aggregating overall blocking and non-blocking opportunities during the before and after data collection, propensities to block of 0.55 and 0.67 were observed, a 12% increase in propensity. Again, similar to the previous intersections, both the before and after propensity rates were high, with drivers in both cases more likely to block than not block. And again, from day to day, significant variations were seen in the likelihood to block.

3.2.6 Clairmont Rd. at I-85 SB, near Sam's Club For the Clairmont Rd. and I-85 SB, near Sam's Club, intersection, 9 hours of propensity data were extracted from both the before and after videos. The 3-hour beforeafter study

44

Operational Evaluation of Do Not Block the Box Campaigns in Georgia

period was 4:007:00 PM. Table 13 provides the number of hours of data extracted for each day, total number of blocking opportunities, total number of observed blocking vehicles, total number of observed non-blocking vehicles, and blocking propensity. The analysis is provided for each day of observation and overall before and after.
There is a significant difference between the Clairmont Rd. and I-85 SB, near Sam's Club intersection propensity data and that of the other intersections. The previous intersections all utilized the blocking definition as presented in Chapter 2. However, for this intersection a vehicle was counted as blocking if it entered the intersection and was unable to immediately exit. Thus, a vehicle that entered the box, stopped, and then was able to exit the box prior to its phase changing was considered to have blocked. In the prior definition, to be considered blocking, a vehicle was required to be in the intersection box when a conflicting movement received its phase change. Similarly, any vehicle that stopped at the stop bar and did not enter the intersection, for any period of time, was counted as not blocking, even if later during the phase indication the vehicle entered the intersection creating a block. In this instance, such vehicle behavior would be counted as both not blocking and blocking.
This broader blocking definition was selected to explore whether the blocking definitions chosen significantly influenced results. Also, this definition captures blocking of permissive movements that is not captured by the previous definition.

Table 13: Clairmont at I-85 SB, near Sam's Club.

Before
Hours of Study Total # Opportunities Total # Blocking

04/11/2016 (Mon) 3
230
159

04/12/2016 (Tue) 3
416
314

04/15/2016 (Fri) 3
381
278

Total
9 1027 751

45

Operational Evaluation of Do Not Block the Box Campaigns in Georgia

Total # Non-Blocking

71

Propensity

0.69

After
Hours of Study Total # Opportunities Total # Blocking Total # Non-Blocking Propensity

04/17/2017 (Mon) 3
361
326
35
0.90

102 0.75
04/18/2017 (Tue) 3 217 191 26 0.88

103 0.73
04/21/2017 (Fri) 3 510 382 128 0.75

276 0.73
9 1088 899 189 0.83

While the number of blocking opportunities under this broader definition is significantly greater, the overall propensity rates are similar to those of the previous intersections, ranging from 0.69 to 0.90. Aggregating overall blocking and non-blocking opportunities during the before and after data collection, propensities to block of 0.73 and 0.83 were observed, a 10% increase in propensity. Again, these results were in line with the previous observations. And once again, in both the before and after data, propensity rates were high, with drivers in both cases more likely to block than not block. However, a reduction in the day-to-day variation was seen.

3.3 Field Result Summary
Table 14 summarizes the overall results of the propensity analysis for all intersections.

46

Operational Evaluation of Do Not Block the Box Campaigns in Georgia

Table 14: Results Summary for All Intersections.

Intersection Name
1. Ashford Dunwoody Rd. NE at Ravinia Dr. NE
2. Peachtree Dunwoody Rd. at Lake Hearn Dr. NE
3. Peachtree Dunwoody Rd. at Abernathy Rd. NE
4. Peachtree Dunwoody Rd. at Johnson Ferry Rd.
5. W. Paces Ferry Rd. NW at I-75 SB On/Off Ramps
6. Clairmont Rd. at I-85 SB, near Sam's Club

Before
After Before After
Before
After
Before After Before After Before After

Total #
Hours 4
9 9
6.5
8
4
10.5 8.75
9 9 9 9

Total # of
Opportunity
28
304 114
44
130
12
219 143 85 33 1027 1088

Total # of
Blocking
20
184 77
39
112
1
182 89 47 22 751 899

Total # of NonBlocking
8
118 37
5
18
11
37 54 38 11 276 189

Prope -nsity
0.71
0.61 0.68
0.89
0.86
Insufficient data 0.83 0.62 0.55 0.67 0.73
0.83

Several characteristics in blocking behavior are evident throughout the analysis. First, the change in propensity between before and after conditions was inconsistent, witnessing both increasing and decreasing propensity. However, regardless of an increase or decrease in blocking rate, the aggregated observed propensities were consistently high in both the before and after conditions. The lowest observed aggregate propensity to block was 55% with all other time periods above 60%, and half of the time periods observed had a propensity to block of 70%. In addition, there was significant variability in day-to-day blocking opportunities. At the same intersection, within the same week there would be peak periods where the opportunity to block was rare or non-existent, while on other days significant blocking opportunities were observed.
47

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
3.4 Simulation Study
As previously discussed, it was not possible to collect operation metrics such as travel time and delay at the intersections. To understand the operational impact of DBTB treatments, a microscopic simulation model of a signalized four-leg intersection that has downstream bottlenecks on its major street was developed. The simulation reflects the propensity of a vehicle entering the intersection box when a blocking opportunity exists and the resulting blocking of traffic should a vehicle "block the box." This study explores the relationship of this blocking behavior to vehicle delay and reduced capacity. Appendix C provides a detailed discussion of the simulation study methodology and findings; thus, only the key insights are provided in the following discussion.
From the delay and capacity reduction results of the simulation study, it is seen that the impact of blocking can be significant, reaching complete gridlock on intersection approaches. However, the results demonstrate that a DBTB treatment can significantly improve flow even without achieving the goal of zero blocking. This is particularly true where blocking propensity is reduced from the mid-range (40% to 60%) to under 20%. From this it may be postulated that many of the intersections that were included in this study could potentially receive a significant benefit if blocking could be reduced to significantly lower rates. While the treatments in the current field study did not demonstrate rate reductions to the level necessary in the simulation study for meaningful operational benefits, the simulation study does highlight the importance for continuing to seek a solution to the DBTB challenge.
48

: CONCLUSIONS AND RECOMMENDATIONS
4.1 Conclusions
Traffic congestion is often an unfortunate reality. As such, it is necessary to manage congestion, minimizing its impacts. In congestion, when a vehicle enters an intersection that has insufficient space to exit on the opposite side due to downstream traffic spillback, it often leads to the obstruction of vehicle and pedestrian movement on conflicting approaches. The effect of "blocking the box" can propagate to nearby intersections and, in extreme cases, lead to gridlock. Along with negative traffic impacts, such as capacity reductions and increased travel times, blocking the box creates potentially unsafe vehicle and pedestrian movement. A "Do Not Block the Box" treatment seeks to reduce the likelihood of drivers entering an intersection when there is insufficient space to exit, and thus reduce blocking occurrences. DBTB treatments are typically low cost, representing a traffic management alternative accessible to most transportation agencies. This report presented a study that explored the performance of DBTB treatments installed in the greater Atlanta, Georgia, area.
For the studied sites, the likelihood, or propensity, of a vehicle to block was measured both before and after the DBTB treatment installation. Several blocking behavior characteristics were observed throughout the analysis. First, the change in propensity with the installation of the DBTB treatment was inconsistent, witnessing both increasing and decreasing blocking rates. However, regardless of an increase or decrease in blocking rate, the aggregated observed propensities at the studied intersections were consistently high in
49

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
both the before and after treatment conditions. The lowest observed aggregate propensity to block was 55%, with all other time periods above 60%, and half of the observed periods with a propensity to block of 70%. In addition, there was significant variability in day-today blocking opportunities. That is, at the same intersection, within the same week, peak periods where opportunities to block were rare or non-existent were observed, as well as peak periods with significant blocking opportunities. Given these findings, it may not be concluded that the installed treatments reliably impacted blocking behavior. In addition, where sites did show improvements, the blocking rate remained high, often well in excess of 50%.
There are potential biases in this study. The first is that the sites were not randomly selected; rather, they were identified as high blocking sites by system managers. These sites may represent the worst-case scenarios and the most difficult to address. In addition, the after data collection at several sites occurred during the I-85 bridge closure. It is not known if the potential rerouting or other driver responses to this incident may have influenced the observed driver behavior. Finally, several sites utilized police officer control during the highest demand periods. If the police were not present, these time periods may have experienced different blocking behavior than the time periods included in the evaluation. However, even given these potential biases, there still remains a failure of the DBTB treatment to address blocking under these conditions.
While not measured at these sites, it is important to highlight that a reduction in blocking may result in significant operational improvements. To explore potential performance impacts of blocking, a microscopic simulation model of blocking behavior was developed. The simulation reflects the propensity of a vehicle entering the intersection
50

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
box when a blocking opportunity exists and the resulting blocking of traffic, should a vehicle "block the box." This model allowed for the exploration of the relationship of blocking behavior to vehicle delay and intersection capacity. The results showed that the impact of blocking can be significant, potentially resulting in gridlock. However, the simulation also was able to demonstrate that DBTB treatments can significantly improve traffic flow even without achieving zero blocking. From this it may be postulated that many of the intersections that were included in this study could significantly benefit if blocking could be reduced. While the treatments in the current field study did not demonstrate the blocking rate reductions necessary for meaningful operational benefits, the simulation study does highlight the importance for continuing to seek a solution to the DBTB challenge.
4.2 Recommendations
Vehicle blocking can significantly impact traffic operations. While the current treatment as a standalone measure did not meaningfully impact blocking behavior, there is significant value in continuing to seek reductions in blocking behavior. Thus, based on the study findings and field observations during the data collection, several recommendations are offered.
1) Signal timing to reduce blocking opportunities. The first strategy to address blocking should be, where possible, the elimination of the potential for blocking utilizing congested-period signal timing that reduces blocking opportunities. Blocking opportunities occur where the flow of vehicles into an intersection exceeds the intersection capacity, often reflected as spillback into upstream
51

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
intersections. Where practical, upstream signal timing should be set to limit downstream vehicle arrivals to that of the downstream intersection processing capacity. While this may result in lower upstream performance, avoiding the gridlock within the network resulting from spillback and blocking should be prioritized. The development of such signal timings will typically require the use of advanced simulation tools applied at the corridor level over time periods greater than the typical peak-hour analysis. While this represents a significant investment in timing plan development, the potential benefits are substantial. 2) Reduction or elimination of free-flow turn movements during congested periods. One key observation at several sites is related to the impact of free-flow turn movements on intersection operations. Under high-demand conditions, a freeflow movement could continuously "fill-in" available capacity on the departure lanes of an intersection approach. This would result in vehicles from a controlled movement utilizing the same departure lanes continually being unable to proceed when they receive a green indication. It is reasonable to hypothesize that this increases the driver frustration and aggressiveness from the controlled movement, resulting in additional blocking, as they saw that as the only opportunity to proceed through the intersection. 3) Limit candidate intersections. While blocking the box occurs when a vehicle stops within the box, in a practical sense many of the observed "blocks" had minimal or no observed impacts on intersection capacity. This could occur for two reasons. The first is that the size of the intersection proper allowed conflicting vehicles to easily maneuver around the blocking vehicle(s). The second is that the
52

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
blocking and blocked vehicles used the same intersection departure lanes. In such an instance, while blocking occurred, there was no intersection capacity impact. In both situations, as drivers saw minimal-to-no benefit of the treatment, this may have increased the likelihood of disregarding the treatment. 4) Public education. A public education program on the benefits of not blocking the box may help to decrease the propensity to block and reinforce the need to follow DBTB treatments. 5) Enforcement. Along with public education, there is likely a need for enforcement. For the given study, none of the intersection DBTB treatments were enforced through citations to drivers that blocked the box. The effectiveness of enforcement in improving the DBTB treatment performance should be explored. This should include different enforcement program durations (i.e., intermittent vs continual), warnings vs. citations, automated vs. manual citations, etc. 6) Additional treatment. Additional treatments should be developed and tested, for instance, flashing DBTB signs that are indicated only when vehicles are detected stopped in intersection departure lanes.
Finally, the most probable means to successfully address intersection blocking is the development of a DBTB overarching program. Such a program would combine the above recommendations into a comprehensive strategy, developing training directed at identifying factors contributing to blocking at specific intersections, developing signal timing plan guidance addressing blocking, applying DBTB signal and striping design, implementing wide-spread education, etc.
53

Operational Evaluation of Do Not Block the Box Campaigns in Georgia 54

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
: REFERENCES
1. Shinar, David. "Safety and Mobility of Vulnerable Road Users: Pedestrians, Bicyclists, and Motorcyclists." Accident Analysis and Prevention 44 (2012): 12. Elsevier Ltd. Web.
2. Centre for International Economics Canberra & Sydney. Business Costs of Traffic Congestion. Rep. Sydney: Centre for International Economics, 2006. Print.
3. Centre for Economics and Business Research. The Future Economic and Environmental Costs of Gridlock in 2030. London, 2014.
4. Popik, B. Entry from April 27, 2005, "Don't Block the Box!" The Big Apple: Don't Block the Box! April 27, 2005. http://www.barrypopik.com/index.php/new_york_city/entry/dont_block_the_box. Accessed July 26, 2016.
5. Harris, D.S. "Traffic Management Alternatives for Business Improvement Districts." In School of Civil and Environmental Engineering, Georgia Institute of Technology, Atlanta, Georgia, 2014, p. 220.
6. Medical Academic and Scientific Community Organization. "Don't Block That Intersection--It Could Cost You!" MASCO. http://www.masco.org/news/spdon%E2%80%99t-block-intersection. Accessed July 26, 2016.
7. Cross, B. "Tickets Adding Up at Austin's Don't Block the Box Intersections." KEYE-TV CBS Austin, August 24, 2015. http://keyetv.com/news/local/ticketsadding-up-at-austin39s-don39t-block-the-box-intersections. Accessed July 26, 2016.
55

Operational Evaluation of Do Not Block the Box Campaigns in Georgia 8. U.S. Department of Transportation, Federal Highway Administration. Manual on Uniform Traffic Control Devices (MUTCD). 2009 Edition, Part 3. pp. 3B-3823B383. http://mutcd.fhwa.dot.gov/htm/2009/part3/fig3b_18_longdesc.htm. Accessed July 29, 2016. 9. Saroj, A., N. Choudhary, H.G. Kim, A. Guin, M. Rodgers, and M. Hunter. "Video Tool for Manually Extracting Complex Traffic Data." TRB 2018 Annual Meeting. http://amonline.trb.org/18-05443-1.3991066?qr=1. Accessed February 9, 2018. 10. Google Maps. 2018. Google Maps. [online] Available at: https://www.google.com.au/maps. Accessed December 2018.
56

Appendix A

ANALYSIS OF VEHICLE BLOCKING BEHAVIOR ON INTERSECTION PERFORMANCE

LIST OF TABLES
Table A-1: Summary of Blocking Laws for Every State, Including the District of Columbia (3, 19)...................................................................................................62
LIST OF FIGURES
Figure A-1: MUTCD "Do Not Block Intersection" Markings (3, 17). ....................61
Figure A-2: Schematic Diagram of the Network Developed in the Simulation Model. ..............................................................................................................................64
Figure A-3: Roadway Sections of the Network on Major Street Approaches to Implement DBTB. ................................................................................................66
Figure A-4: Operation of the Priority Rule to Incorporate the Non-blocking Behavior for Vehicle Type 3................................................................................................68
Figure A-5: Simulation Run Snapshot (60% Blocking Likelihood) Showing Dynamic Assignment of Vehicle Type with Change in Vehicle Color. ..............................69
Figure A-6: Logic Flowchart of the DBTB Dynamic Assignment of Vehicle Type. .....................................................................................................................70
Figure A-7: (a) Scatterplot of Average Delay (sec/veh) on Minor Street Versus Blocking Likelihood, and (b) Boxplot of Average Delay (sec/veh) on Minor Street Versus Blocking Likelihood.................................................................................74
Figure A-8: (a) Scatterplot of Max 15-min Average Delay (sec/veh) on Minor Street Versus Blocking Likelihood, and (b) Boxplot of Maximum 15-min Average Delay on Minor Street Vehicle Versus Blocking Likelihood. ........................................75
Figure A-9: Scatterplot Depicting Variability in the Minor Street Capacity Loss Across Different Levels of Blocking Likelihoods. ..............................................77

57

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
A.1 Abstract
While some congestion may be inevitable, it is critical that congestion not escalate due to gridlock. As such, "Do Not Block the Box" (DBTB) treatments represent a potential low cost, traffic operations mitigation measure aimed at minimizing gridlock in congested intersection networks. A vehicle is considered to "Block the Box" when it enters the intersection with no space to exit due to the traffic spillback downstream. While the use of a DBTB treatment as a control strategy has been adopted by many cities in the U.S., there is a lack of literature on its effectiveness. This research explores the performance of DBTB treatments by quantifying the impact of vehicle blocking behavior on intersection performance. For this, a microscopic simulation model of a signalized four-leg intersection that has downstream bottlenecks on its major street was developed. The simulation reflects the likelihood of a vehicle entering the intersection box when a blocking opportunity exists and the resulting blocking of traffic should a vehicle "block the box." This study explores the relationship of this blocking behavior to vehicle delay and reduced capacity. From the delay and capacity-reduction results, it is seen that the impact of blocking can be significant, reaching complete gridlock on intersection approaches. Ultimately, the goal of a DBTB treatment is to reduce the number of blocking events to zero. However, from the result of this study it is seen that a DBTB treatment can significantly improve flow even without achieving the goal of zero blocking.
Keywords: Do Not Block the Box, Public Information, Gridlock, Traffic Enforcement, Traffic Simulation
58

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
A.2 Introduction
Traffic congestion negatively impacts quality of life, hampers business activities, and adds to harmful vehicular emissions (1). In 2013 alone, traffic congestion cost the U.S. an estimated $124B, a value that is projected to increase 50% by 2030 (2). While some congestion may be inevitable, it is critical that congestion not escalate due to gridlock. As such, "Do Not Block Intersection" or "Do Not Block the Box" treatments represent a potential low cost, traffic operations mitigation measure aimed at minimizing gridlock in congested intersection networks. DBTB treatments have been successfully implemented in several parts of the world and represent a traffic management alternative available to federal, state, and local transportation agencies and groups (3). This paper explores the potential impact of DBTB treatments at congested signalized intersections.
"Blocking the box" occurs when a vehicle with right-of-way (e.g., a green indication at a signalized intersection) enters the intersection with insufficient space to exit on the opposite side due to downstream traffic spillback. This vehicle must then stop within the intersection proper, or "box," potentially obstructing the movement of pedestrians and vehicles with right-of-way on conflicting approaches. For instance, blocking a permitted turn movement during the current phase or blocking cross-street traffic if the vehicle remains trapped in the intersection after the current phase terminates. The compounding of multiple blocking events on a congested network can lead to gridlock situations and excessive delays (3). A DBTB treatment seeks to reduce the likelihood of drivers entering an intersection when there is insufficient space to exit the box, and thus reduce blocking occurrences and the potential for gridlock.
59

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
A.3 Background
Characteristics of gridlock and strategies to control it has long been a topic of interest to researchers (4, 5, 6, 7, 8). It is been shown that avoiding growth of small localized gridlock can prevent a network-level gridlock ("jam" state) (9). However, there is lack of detailed study on DBTB treatments performance and efficiency (3). One of the few studies identified considered the DBTB treatment implemented by the Boston Transportation Department in partnership with the Medical Academic and Scientific Community Organization (MASCO) and the Boston Police Department. This study observed a 50% decrease in intersection blocking and it reported a reduction of 22% to 64% in number of citations after enforcing DBTB treatments (3, 10). A number of additional studies have also reported on citations and warnings related to DBTB enforcement (3, 11, 12, 13, 14); however, while the number of blocking events could be quantified by tracking citations the driver time savings due to this DBTB and enforcement must be estimated through other means.
While studies on DBTB operations are limited, the history of using DBTB treatments to avert gridlock dates back to at least 1964, with success of the first recorded DBTB treatment, installed in London, England. In the 1970s, the first U.S. DBTB treatments were installed in New York. Since those first installments, DBTB programs have increased in popularity as a gridlock mitigation measure (3, 15). As of today, DBTB has been adopted in many U.S. cities, including Boston, Miami, Austin, San Francisco, Atlanta, etc.
Installing DBTB at an intersection is a simple and low-cost process. It consists of implementing a striping treatment adhering to the standards in the Manual on Uniform Traffic Control Devices (MUTCD), as shown in Figure A-1. In addition, signs stating "Do
60

Operational Evaluation of Do Not Block the Box Campaigns in Georgia Not Block Intersection" or "Do Not Block the Box" are installed near the intersection. This striping treatment visually warns the driver to avoid queueing within the intersection box. The cost of installing a DBTB treatment involves cost of painting the pavement markings and installing the signs and typically ranges from $1000 to $2000 with a comparable 20year maintenance cost. A significant share of DBTB program cost goes to enforcement, which is done either through parking attendants, police, or automated gridlock cameras (3). For instance, within the StreetSafe initiative launched in 2013, the Washington D.C. Metropolitan Police Department (MPD) installed gridlock cameras at 20 intersections (16).
Figure A-1: MUTCD "Do Not Block Intersection" Markings (3, 17). Drivers' adherence to DBTB regulations is crucial for the success of DBTB treatments. U.S. traffic codes that enable issuing citations to block-the-box violators,
61

Operational Evaluation of Do Not Block the Box Campaigns in Georgia

generally fall into three categories: 1) obstructing, 2) stopping, and 3) sign laws. The obstructing laws prohibit a driver from entering an intersection that has insufficient space to exit, regardless of the traffic control signal indication. The stopping laws prohibit a vehicle from standing, stopping, or parking within an intersection, unless necessary either to avoid conflict or comply with the directions of a police officer or traffic control device. Lastly, in some states, the sign laws reinforce the stopping laws, restricting the driver from stopping at posted locations (3). While most jurisdictions consider blocking-the-box as a moving violation, some cities (e.g., New York) classify it as non-moving to enable both police officers and parking attendants to issue citations thus greatly increasing the number of people that may enforce DBTB (3, 18). Table A-1 summarizes presence of DBTB laws in state traffic codes.

Table A-1: Summary of Blocking Laws for Every State, Including the District of Columbia (3, 19).

State
Alabama, California, Colorado, Georgia, Hawaii, Minnesota, North Dakota, Ohio, Pennsylvania, Washington
Delaware, District of Columbia, Florida, Idaho, Michigan, Nevada, New York, Oregon, South Carolina Arizona, Arkansas, Illinois, Kansas, Louisiana, Maryland, Mississippi, Nebraska, New Hampshire, New Mexico, Oklahoma, Rhode Island, South Dakota, Tennessee, Texas, Vermont, West Virginia, Wyoming Alaska, Missouri, New Jersey, North Carolina, Utah, Virginia
Montana
Connecticut, Indiana, Iowa, Kentucky, Maine, Massachusetts, Wisconsin

Obstructing Law
-

Stopping Law
-

Sign Law
-

62

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
A.4 Methodology
A traffic simulation model consisting of a signalized four-leg intersection, a six-lane major arterial crossing a four-lane minor street, developed in VISSIM, was employed to investigate DBTB. To generate blocking opportunities at the intersection, downstream bottlenecks are placed on the major arterial, creating spillback (i.e., queuing) through the intersection box. Major street vehicles may follow either blocking (i.e., will enter the intersection box when the exit is blocked by a queue) or non-blocking (i.e., will not enter an intersection if a block would result) behavior. The selected behavior is determined randomly according to a user defined blocking likelihood, i.e., the likelihood that a vehicle will exhibit blocking behavior. For instance, a blocking likelihood of "zero" precludes any vehicle from entering the intersection if the entry would cause a block to occur, while a blocking likelihood of "one" indicates that all vehicles will enter the intersection without concern for the potential to create a block. This modeling approach allows for the exploration of the sensitivity of intersection operations to different levels of blocking likelihood and to the impact of a reduced likelihood due to a DBTB treatment. Implementation of this methodology is described in the remainder of this section.
A.4.1 PTV VISSIM Traffic Simulation Software The traffic simulation used in this research was implemented in PTV-VISSIM 5.2 a commercially available microscopic transportation simulation package. In this model, traffic flow is based on the Wiedemann's car-following models and rule-based algorithms for lateral vehicle movement (20, 21). This effort required use of the VISSIM COM (component object model) interface that allows access to the object model hierarchy, with network elements such as vehicles, links, vehicle inputs, etc. (22).
63

Operational Evaluation of Do Not Block the Box Campaigns in Georgia A.4.2 Network Layout
The simulation model is shown schematically in Figure A-2: . The left image is a snapshot of the network and the right image is a sketch of the signal layout. Signal A controls the traffic movement at the Major Street and Minor Street intersection. Signals B and C are placed on the major street, downstream of the intersection, to function as traffic bottlenecks.
Figure A-2: Schematic Diagram of the Network Developed in the Simulation Model. A.4.3 Implementing Blocking Behavior
To simulate a blocking incident a vehicle must enter the intersection when a blocking opportunity exists. That is, traffic spillback (i.e., queuing) from a downstream bottleneck must reach the subject intersection, leaving insufficient space for a vehicle that enters the intersection box to exit. To generate the spillback in this study, fixed-time signal phase lengths were chosen such that the Signal B (or Signal C) hourly capacity is less than that of the upstream Signal A approach. Thus, as the mainline flow increased the capacity of
64

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Signal B (or Signal C) would be exceeded prior to that of Signal A, allowing for the development of a queue between Signal B (or Signal C) and Intersection A.
At several intersections in Atlanta it was observed that not all drivers choose to enter an intersection box when that action could result in blocking. Thus, it is also necessary that the simulation reflects the likelihood of a vehicle entering the intersection box when a blocking opportunity exists. For this effort, this is referred to as the blocking likelihood. To implement blocking likelihood in VISSIM, dynamic assignment of the Vehicle Type attribute of the Vehicle Object is utilized in coordination with Priority Rules.
A.4.4 Dynamic Assignment of Vehicle Type Three different vehicle types are defined:
Vehicle Type 1 This vehicle has the default characteristics. All vehicles enter the simulation as a type 1 vehicle.
Vehicle Type 2 Vehicle type 1 with driver behavior to enter an intersection box irrespective of space availability to exit the box. That is, a vehicle that can create a blocking event.
Vehicle Type 3 Vehicle type 1 with driver behavior that will not enter the intersection when insufficient space exists at the box exit. That is, a vehicle that will not create a blocking event.
To implement dynamic assignment of vehicle type according to the blocking likelihood each major street approach of the DBTB intersection is divided into two sections: 1) the decision-zone, i.e., the area where vehicles are assigned as vehicle type 2 or vehicle type 3; and 2) the box zone, i.e., the area including the intersection proper and
65

Operational Evaluation of Do Not Block the Box Campaigns in Georgia one vehicle length downstream. The schematic representation of the sections is shown in Figure A-3: .
Figure A-3: Roadway Sections of the Network on Major Street Approaches to Implement DBTB.
Dynamic assignment of the vehicle type is implemented in VB.Net using the VISSIM COM interface. The first step is to determine if the potential for blocking exists. This step acts only as a filter, eliminating the need to dynamically assign vehicle type when blocking is not possible. While elimination of this filter would not alter the simulation results, it serves to reduce internal calculations required to assign vehicle type and significantly enhance the execution speed of the simulation. In the current implementation, a vehicle speed of 15 mph in the box zone was used for this filter.
In the second step of the dynamic assignment process, applied only when the potential for blocking has been indicated, COM is used to identify vehicles within the decision zone using the approach link number and vehicle coordinate attributes. The vehicle closest to the stop line in the decision-zone is assigned as the lead vehicle. The lead vehicle is then assigned as vehicle type 2 with a probability of blocking likelihood; otherwise, the lead vehicle is assigned as vehicle type 3. Those vehicles in decision zone upstream of the leading vehicle are then designated as following vehicles and assigned the same vehicle type as the lead vehicle. The vehicle type assignments for the lead and following vehicles
66

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
is undertaken on a lane-by-lane basis. The assignment of the following vehicle behavior to that of the lead vehicle is based on observation of blocking behavior in Atlanta, Georgia. It was observed that when a vehicle made a decision to block, several vehicles behind that vehicle (i.e., following vehicles) also had a very high tendency to enter the intersection. Future efforts will seek to formalize the relationship between the lead vehicle blocking likelihood and the subsequent following vehicles blocking likelihood.
A.4.5 Implementing Blocking Rules VISSIM priority rules are used to enable the conditional stopping necessary to implement the desired blocking and non-blocking behavior based on the assigned vehicle type. Priority rules are based on the headway and gap conditions of a vehicle at specified location and are specific to a vehicle type. Two elements are required to implement a priority rule in VISSIM: 1) a stop line, and 2) one or more conflict markers associated with the stop line. In this implementation, the stop line of the priority rule is placed at the stop bar of the major street approach, with the conflict marker placed one vehicle length downstream of the intersection box. Figure A-4 shows this framework in the model. The minimum headway, defined as the length of the conflict area, is set to extend from the conflict marker to the approach stop bar (yellow area in Figure A-4: ). The minimum gap, defined as the time until a conflicting vehicle reaches the conflict marker, is set to two seconds. Thus, if a vehicle is within the area between the stop bar and the conflict marker or within two seconds of the conflict marker, the priority rule will be active and an adhering vehicle type will not enter the intersection box. Finally, the priority rule was conditioned on the speed of the subject vehicle type, with approximately 18 mph for the priority rule to apply. In this implementation, only vehicle type 3 adheres to the priority rule. Thus, vehicle type
67

Operational Evaluation of Do Not Block the Box Campaigns in Georgia 3 will exhibit non-blocking behavior, while vehicle type 2 will enter an intersection box when the possibility of creating a block exists.
Figure A-4: Operation of the Priority Rule to Incorporate the Non-blocking Behavior for Vehicle Type 3.
Figure A-5: is a snapshot of the simulation with a blocking likelihood of 60%. The image shows vehicle type indicated by color. A change in vehicle type associated with the dynamic assignment of vehicles of vehicle type 1 is seen on every lane of the major street. For example, considering Eastbound traffic, the yellow colored EB vehicles of vehicle type 1 change to vehicle type 2 or vehicle type 3 in the decision zone. Furthermore, vehicles of vehicle type 2 on the two leftmost lanes of EB approach block the vehicles on the NB approach while the vehicles of vehicle type 3 on the rightmost lane exhibit non-blocking behavior and remain out of the box.
68

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure A-5: Simulation Run Snapshot (60% Blocking Likelihood) Showing Dynamic Assignment of Vehicle Type with Change in Vehicle Color.
Figure A-6: displays the flowchart of the simulation COM logic for the dynamic assignment of vehicle type. This logic is implemented in VB.NET and executed each simulation time step, for each Major Street approach lane.
69

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure A-6: Logic Flowchart of the DBTB Dynamic Assignment of Vehicle Type. 70

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
A.5 Experimental Design
The base simulation scenario consisted of a 3-hour simulation allowing for the collection of performance metrics during major street near-capacity conditions, overcapacity conditions, where blocking could occur, and a recovery period. To accomplish this the firsthour major street traffic demand was set just below the capacity of the downstream Signals B and C, the second-hour traffic demand was set over the capacity of Signals B and C while under the capacity of the Intersection A major street approaches, and the third hour major street traffic demand was set under the capacity of Signals B and C. The minor street had consistent traffic demand throughout the simulation run. All signals were fixed time. Signals B and C near-capacity (1900 vph) and over-capacity (2600 vph) demands were determined through iterative runs on the base network.
Simulation experiments were conducted to model the delay incurred and the reduction in the number of vehicles processed on the minor street approaches for three under-saturated traffic conditions (100, 200, and 300 vph) as well as oversaturated conditions (standing minor street approach queue throughout the simulation), under various blocking likelihoods (0%, 20%, 40%, 60%, 80%, and 100%). The reduction in the number of vehicles processed when the minor street was oversaturated also represents reduction in minor street capacity during blocking. In all scenarios, only through vehicles are modeled. Ten replicate trials were conducted for each traffic demand with blocking likelihood combination.
71

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
A.6 Results
The focus of the simulation results is on the minor street vehicles, since the given scenarios' blocking has minimal impact on the major street performance. As there are no turning movements a major street vehicle choosing not to block does not compete for space with minor street vehicles turning onto the major street. While the addition of turning movements will allow for the capture of additional interactions, it is not expected to change the overall observed trends.
A.6.1 Delay Figure A-7: (a) and Figure A-7: (b) show the scatterplot and box plots, respectively,
for the average delay (sec/veh) over the 3-hour run, obtained for the 100 vph, 200 vph, and 300 vph minor street volumes across blocking likelihoods. For this analysis, the average vehicle delay was determined for every five-minute interval and the reported average delay is the average of these intervals.
As expected, as the minor street volume increases, the delay values also increase. The blocking likelihood of 0% shows the expected delay for no blocking by major street vehicles. As the blocking likelihood increases the delay increases, with the most significant increases in mean delay and variability at 40% and 60%. The higher minor street volumes also have more dramatic increases in delay and variability, as seen in the box plots. These values may be conservative, as the simulation does not reflect that these increasing delays may increase minor street vehicle aggressiveness, resulting in additional blocking as minor street vehicles attempt to force their traversal of the intersection. Figure A-8: (a) and Figure A-8: (b) show the scatterplot and box plots, respectively, for the maximum 15-minute average delay (sec/veh). These delay values represent the worst case performance
72

Operational Evaluation of Do Not Block the Box Campaigns in Georgia experienced by vehicles, with instances at the highest blocking likelihoods representing complete gridlock with delays approaching one hour. Although the range of average delay values obtained in Figure A-8: are more than twice that shown in Figure A-7: the basic pattern in variation across different blocking likelihood and minor street volumes remains similar.
73

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
(a)
(b) Figure A-7: (a) Scatterplot of Average Delay (sec/veh) on Minor Street Versus Blocking Likelihood, and (b) Boxplot of Average Delay (sec/veh) on Minor Street
Versus Blocking Likelihood. 74

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
(a)
(b) Figure A-8: (a) Scatterplot of Max 15-min Average Delay (sec/veh) on Minor Street Versus Blocking Likelihood, and (b) Boxplot of Maximum 15-min Average Delay on
Minor Street Vehicle Versus Blocking Likelihood. 75

Operational Evaluation of Do Not Block the Box Campaigns in Georgia A.6.2 Minor Street Capacity Reduction
The results obtained for the oversaturated demand on the side street are shown in Figure A-9:. The scatterplot shows the reduction in minor street processed traffic due to blocking on the major street. Given the oversaturated conditions, (i.e., there was a continuous standing queue) minor street processed traffic was approximately 2400 vehicles during the 3-hour simulation run (i.e., an hourly capacity of approximately 800 veh/hr), determined as the number of minor street vehicles to traverse the intersection when no major street vehicles were blocking, i.e. blocking likelihood of zero. The reduction in traffic processed in Figure A-9: represents how many fewer vehicles departed the Minor Street approach during the 3-hour run, for the varying likelihoods, with nearly all reductions occurring in the second hour during blocking.
The scatterplots again indicate that the most dramatic reductions in traffic processed occur in the mid-range blocking likelihoods of 40% and 60%, with reductions due to blocking equivalent to 60% to 100% of an hour of capacity. Also, agreeing with the prior delay results, nearly complete gridlock is seen in the 80% and 100% blocking likelihoods with reductions approaching and exceeding the capacity of the entire second hour of the Major Street blocking period.
76

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure A-9: Scatterplot Depicting Variability in the Minor Street Capacity Loss Across Different Levels of Blocking Likelihoods.
A.7 Conclusion
This study explores the relationship between blocking behavior, increased vehicle delay, and capacity reduction in a single intersection scenario. From the delay and capacity reduction results, it is seen that the impact of blocking can be significant, reaching complete gridlock on intersection approaches. Ultimately, the goal of a DBTB treatment is to reduce the blocking likelihood to zero, or nearly so. However, from the results it can be seen that a DBTB treatment can significantly improve flow even without achieving the goal of zero blocking. This is particularly true where blocking likelihood is reduced from the mid-range (40% to 60%) to under 20%. This also demonstrates the importance of enforcement
77

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
programs. While it is not necessary that enforcement eliminate blocking altogether, they must be of sufficient frequency to limit those drivers willing to risk blocking to a low percentage of the driving population.
While the results highlight the potentially significant impact of blocking, and the improvements that could be achieved through DBTB treatments, several challenges remain in the analysis. The first is regarding model validation. Current validation is limited to observational comparisons with in-field DBTB treatments. However, ongoing data collection efforts are underway to quantify before-and-after DBTB treatment operations, allowing for further model calibration and validation. In addition, the models could be expanded to include multiple intersections (directly capturing gridlock between intersections) and turning movements to reflect potential additional interaction between the cross streets.

A.8 References

1.

Weisbrod, G., D. Vary, and G. Treyz. Measuring the Economic Costs of Urban Traffic

Congestion to Business. In Transportation Research Record: Journal of the

Transportation Research Board, No.1839(1), Transportation Research Board of National

Academies, Washington, D.C., 2003, p. 98106.

2.

Centre for Economics and Business Research. The Future Economic and Environmental

Costs of Gridlock in 2030. London, 2014.

3.

Harris, D.S. Traffic Management Alternatives For Business Improvement Districts. In

School of Civil and Environmental Engineering, Georgia Institute of Technology, Atlanta,

Georgia, 2014, p. 220.

78

Operational Evaluation of Do Not Block the Box Campaigns in Georgia

4.

Daganzo, C.F. Improving City Mobility through Gridlock Control: An Approach and Some

Ideas. 2005.

5.

Daganzo, C.F. Urban Gridlock: Macroscopic Modeling and Mitigation Approaches.

Transportation Research Part B: Methodological, 2007. Vol. 41(1): p. 4962.

6.

Lo, H., and A. Chow. Control Strategies for Oversaturated Traffic. Journal of

Transportation Engineering, 2004. Vol. 130(4): p. 466478.

7.

Dinopoulou, V., C. Diakaki, and M. Papageorgiou. Applications of the Urban Traffic

Control Strategy TUC. European Journal of Operational Research, 2006. Vol. 175(3): p.

16521665.

8.

El Faouzi, N.-E., C. Osario, X. Chen, M. Marsico, M. Talas, J. Gao, and S. Zhang. 4th

International Symposium of Transport Simulation (ISTS'14) Selected Proceedings,

Ajaccio, France, 14 June 2014 Reducing Gridlock Probabilities via Simulation-based

Signal Control. Transportation Research Procedia, 2015. Vol. 6: p. 101110.

9.

Hoogendoorn, S. P. , V. L. Knoop, H. Lint, H. S. Mahmassani, M. Saberi, and K. A.

Zockaie. 20th International Symposium on Transportation and Traffic Theory (ISTTT

2013) Urban Network Gridlock: Theory, Characteristics, and Dynamics. Procedia Social

and Behavioral Sciences, 2013. Vol. 80: p. 7998.

10. Medical Academic and Scientific Community Organization. Don't Block That

Intersection--It Could Cost You!. MASCO. http://www.masco.org/news/sp-

don%E2%80%99t-block-intersection. Accessed July 26, 2016.

11. Lozano, Z. More than 1,100 Ticketed in 3 Weeks for `Blocking the Box'. KXAN.com, May

7, 2015. http://kxan.com/2015/05/07/more-than-1100-ticketed-in-3-weeks-for-blocking-

the-box/. Accessed April 22, 2016.

12. Whitehead, T. Most Lucrative Box Junction Rakes in Millions from Unsuspecting Drivers.

The Daily Telegraph, January 13, 2013. http://www.telegraph.co.uk/news/uknews/law-

79

Operational Evaluation of Do Not Block the Box Campaigns in Georgia

and-order/9798763/Most-lucrative-box-junction-rakes-in-millions-from-unsuspecting-

drivers.html. Accessed July 29, 2016.

13. Levy, A. Moneybox Junction: The West London Traffic Light 'Trap' that Rakes in 2.7m

a Year. Daily Mail Online, January 13, 2013. http://www.dailymail.co.uk/news/article-

2261734/Moneybox-junction-The-West-London-traffic-light-trap-rakes-2-7m-year.html.

Accessed July 29, 2016.

14. Cross, B. Tickets Adding Up at Austin's Don't Block the Box Intersections. KEYE-TV CBS

Austin, August 24, 2015. http://keyetv.com/news/local/tickets-adding-up-at-austin39s-

don39t-block-the-box-intersections. Accessed July 26, 2016.

15. Popik, B. Entry from April 27, 2005 Don't Block the Box!. The Big Apple: Don't Block

the

Box!,

April

27,

2005.

http://www.barrypopik.com/index.php/new_york_city/entry/dont_block_the_box.

Accessed July 26, 2016.

16. Segraves, M. D.C. Police Adding Cameras to Enforce Stop Signs, Blocking the Box,

Yielding to Pedestrians. NBC Washington, August 29, 2013.

http://www.nbcwashington.com/news/local/DC-Police-Adding-Cameras-to-Enforce-

Stop-Signs-Blocking-the-Box-Yielding-to-Pedestrians-221531301.html. Accessed July

29, 2016.

17. U.S. Department of Transportation, Federal Highway Administration. Manual on Unifrom

Traffic Control Devices (MUTCD). 2009 Edition, Part 3. p. 3B-3823B-383.

http://mutcd.fhwa.dot.gov/htm/2009/part3/fig3b_18_longdesc.htm. Accessed July 29,

2016.

18. Cardwell, D. Mayor Has New Plan to Cite Drivers Who Block the Box. The New York

Times, May 25, 2007. http://www.nytimes.com/2007/05/25/nyregion/25traffic.html?_r=0.

Accessed July 26, 2016.

80

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
19. US Law, Case Law, Codes, Statutes & Regulations Justia Law. http://law.justia.com/. Accessed April 4, 2014.
20. Fellendorf, M., and P. Vortisch. Microscopic Traffic Flow Simulator VISSIM. In Fundamentals of Traffic Simulation, J. Barcel, Editor. 2010, Springer New York: New York, NY. p. 6393.
21. PTV Vision, VISSIM 5.20 User Manual. 2009. 22. PTV Vision, VISSIM 5.20-10 COM Interface Manual. 2010. p. 262.
81

Appendix B STUDY SITES
LIST OF TABLES
Table B-1: List of Intersections. ...............................................................................89 Table B-2: Hours of Video Data Collected and Processed.......................................90
LIST OF FIGURES
Figure B-1: Parkside Pl. & Perimeter Summit Pkwy., Source: Google Street View. ..............................................................................................................................91
Figure B-2: Parkside Pl. & Perimeter Summit Pkwy., Video Capture, Location 1 10/23/2015 (Before). ............................................................................................91
Figure B-3: Parkside Pl. & Perimeter Summit Pkwy., Video Capture, Location 2 10/23/2015 (Before). ............................................................................................92
Figure B-4: Ashford Dunwoody Rd. NE & Ravinia Dr. NE, Source: Google Street View. ....................................................................................................................92
Figure B-5: Ashford Dunwoody Rd. NE & Ravinia Dr. NE, Video Capture, Location 1 10/26/2015 (Before). .......................................................................93
Figure B-6: Ashford Dunwoody Rd. NE & Ravinia Dr. NE, Video Capture, Location 2 10/26/2015 (Before). .......................................................................93
Figure B-7: Ashford Dunwoody Rd. NE & Ravinia Dr. NE, Video Capture, Location 3 10/26/2015 (Before). .......................................................................94
Figure B-8: Ashford Dunwoody Rd. NE & Ravinia Dr. NE, Video Capture, Location 4 10/26/2015 (Before). .......................................................................94
Figure B-9: Ashford Dunwoody Rd. NE & Ravinia Dr. NE, Video Capture, Location 1 05/06/2016 (After). .........................................................................95
Figure B-10: Ashford Dunwoody Rd. NE & Ravinia Dr. NE, Video Capture, Location 2 04/19/2016 (After). .........................................................................95
83

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-11: Ashford Dunwoody Rd. NE & Ravinia Dr. NE, Before DBTB Lane Configuration and Camera Location. ...................................................................96
Figure B-12: Ashford Dunwoody Rd. NE & Ravinia Dr. NE, After DBTB Lane Configuration and Camera Location. ...................................................................96
Figure B-13: Peachtree Dunwoody Rd. & Lake Hearn Dr. NE, Source: Google Street View. ..........................................................................................................97
Figure B-14: Peachtree Dunwoody Rd. & Lake Hearn Dr. NE, Video Capture, Location 1 10/23/2015 (Before). .......................................................................97
Figure B-15: Peachtree Dunwoody Rd. & Lake Hearn Dr. NE, Video Capture, Location 2 10/23/2015 (Before). .......................................................................98
Figure B-16: Peachtree Dunwoody Rd. & Lake Hearn Dr. NE, Video Capture, Location 3 10/23/2015 (Before). .......................................................................98
Figure B-17: Peachtree Dunwoody Rd. & Lake Hearn Dr. NE, Video Capture, Location 1 05/06/2016 (After). .........................................................................99
Figure B-18: Peachtree Dunwoody Rd. & Lake Hearn Dr. NE, Video Capture, Location 2 05/06/2016 (After). .........................................................................99
Figure B-19: Peachtree Dunwoody Rd. & Lake Hearn Dr. NE, Before Lane Configuration and Camera Location. .................................................................100
Figure B-20: Peachtree Dunwoody Rd. & Lake Hearn Dr. NE, After Lane Configuration and Camera Location. .................................................................100
Figure B-21: Peachtree Dunwoody Rd. & Johnson Ferry Rd., Source: Google Street View. ..................................................................................................................101
Figure B-22: Peachtree Dunwoody Rd. & Johnson Ferry Rd., Video Capture, Location 1 10/26/2015 (Before). .....................................................................101
Figure B-23: Peachtree Dunwoody Rd. & Johnson Ferry Rd., Video Capture, Location 2 10/26/2015 (Before). .....................................................................102
Figure B-24: Peachtree Dunwoody Rd. & Johnson Ferry Rd., Video Capture, Location 3 10/26/2015 (Before). .....................................................................102
Figure B-25: Peachtree Dunwoody Rd. & Johnson Ferry Rd., Video Capture, Location 4 10/26/2015 (Before). .....................................................................103
Figure B-26: Peachtree Dunwoody Rd. & Johnson Ferry Rd., Video Capture, Location 1 04/25/2016 (After). .......................................................................103
84

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-27: Peachtree Dunwoody Rd. & Johnson Ferry Rd., Video Capture, Location 2 04/25/2016 (After). .......................................................................104
Figure B-28: Peachtree Dunwoody Rd. & Johnson Ferry Rd., Before Lane Configuration and Camera Location. .................................................................104
Figure B-29: Peachtree Dunwoody Rd. & Johnson Ferry Rd., After Lane Configuration and Camera Location. .................................................................105
Figure B-30: Peachtree Dunwoody Rd. & Abernathy Rd. NE, Source: Google Street View. ........................................................................................................105
Figure B-31: Peachtree Dunwoody Rd. & Abernathy Rd. NE, Video Capture, Location 1 10/28/2015 (Before). .....................................................................106
Figure B-32: Peachtree Dunwoody Rd. & Abernathy Rd. NE, Video Capture, Location 2 10/28/2015 (Before). .....................................................................106
Figure B-33: Peachtree Dunwoody Rd. & Abernathy Rd. NE, Video Capture, Location 3 10/28/2015 (Before). .....................................................................107
Figure B-34: Peachtree Dunwoody Rd. & Abernathy Rd. NE, Video Capture, Location 4 10/28/2015 (Before). .....................................................................107
Figure B-35: Peachtree Dunwoody Rd. & Abernathy Rd. NE, Video Capture, Location 1 04/18/2016 (After). .......................................................................108
Figure B-36: Peachtree Dunwoody Rd. & Abernathy Rd. NE, Video Capture, Location 2 04/18/2016 (After). .......................................................................108
Figure B-37: Peachtree Dunwoody Rd. & Abernathy Rd. NE, Before Lane Configuration and Camera Location. .................................................................109
Figure B-38: Peachtree Dunwoody Rd. & Abernathy Rd. NE, After Lane Configuration and Camera Location. .................................................................109
Figure B-39: Peachtree Rd. NE & Lenox Rd. NE, Source: Google Street View. ..................................................................................................................110
Figure B-40: Peachtree Rd. NE & Lenox Rd. NE, Video Capture, Location 1 05/02/2016 (Before). ..........................................................................................110
Figure B-41: Peachtree Rd. NE & Highland Dr. NE, Source: Google Street View. ..................................................................................................................111
Figure B-42: Peachtree Rd. NE & Highland Dr. NE, Video Capture, Location 1 05/02/2016 (Before). ..........................................................................................111
85

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-43: Peachtree Rd. NE & Stratford Rd. NE, Source: Google Street View. ..................................................................................................................112
Figure B-44: Peachtree Rd. NE & Stratford Rd. NE, Video Capture, Location 1 05/02/2016 (Before). ..........................................................................................112
Figure B-45: Peachtree Rd. NE & Stratford Rd. NE, Video Capture, Location 1 04/19/2017 (After)..............................................................................................113
Figure B-46: Peachtree Rd. NE & Stratford Rd. NE, Video Capture, Location 2 04/19/2017 (After)..............................................................................................113
Figure B-47: Ponce De Leon Ave. at City Hall East, Source: Google Street View. ..................................................................................................................114
Figure B-48: Ponce De Leon Ave. at City Hall East, Video Capture, Location 1 03/18/2016 (Before). ..........................................................................................114
Figure B-49: S. Cobb Dr. & Pearl St., Source: Google Street View. ..................115
Figure B-50: S. Cobb Dr. & Pearl St., Video Capture, Location 1 03/21/2016 (Before). .............................................................................................................115
Figure B-51: S. Cobb Dr. & Pearl St., Video Capture, Location 2 03/21/2016 (Before). .............................................................................................................116
Figure B-52: S. Cobb Dr. & Walker St., Source: Google Street View................116
Figure B-53: S. Cobb Dr. & Walker St., Video Capture, Location 1 03/21/2016 (Before). .............................................................................................................117
Figure B-54: S. Cobb Dr. & Walker St., Video Capture, Location 2 03/21/2016 (Before). .............................................................................................................117
Figure B-55: W. Paces Ferry Rd. NW & I-75 SB On/Off Ramps, Source: Google Street View. ........................................................................................................118
Figure B-56: W. Paces Ferry Rd. NW & I-75 SB On/Off Ramps, Video Capture, Location 1 03/28/2016 (Before). .....................................................................118
Figure B-57: W. Paces Ferry Rd. NW & I-75 SB On/Off Ramps, Video Capture, Location 2 03/28/2016 (Before). .....................................................................119
Figure B-58: W. Paces Ferry Rd. NW & I-75 SB On/Off Ramps, Video Capture, Location 1 04/04/2017 (After). .......................................................................119
Figure B-59: W. Paces Ferry Rd. NW & I-75 SB On/Off Ramps, Video Capture, Location 2 04/04/2017 (After). .......................................................................120
86

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-60: W. Paces Ferry Rd. NW & I-75 SB On/Off Ramps, Before and After Lane Configuration and Camera Location. ........................................................120
Figure B-61: Mount Paran Rd. NW at I-75 NB Off Ramp, Source: Google Street View. ..................................................................................................................121
Figure B-62: Mount Paran Rd. NW at I-75 NB Off Ramp, Video Capture, Location 1 03/28/2016 (Before). .......................................................................................121
Figure B-63: Mount Paran Rd. NW at I-75 NB Off Ramp, Video Capture, Location 2 03/28/2016 (Before). .......................................................................................122
Figure B-64: Williams St. NW & 10th St. NW, Source: Google Street View. ...122
Figure B-65: Williams St. NW & 10th St. NW, Video Capture, Location 1 04/06/2016 (Before). ..........................................................................................123
Figure B-66: Williams St. NW & 10th St. NW, Video Capture, Location 2 04/06/2016 (Before). ..........................................................................................123
Figure B-67: Williams St. NW & 10th St. NW, Video Capture, Location 1 05/09/2017 (After)..............................................................................................124
Figure B-68: Williams St. NW & 10th St. NW, Video Capture, Location 2 05/09/2017 (After)..............................................................................................124
Figure B-69: South Cobb Dr. & I-285 SB On/Off Ramps, Source: Google Street View. ..................................................................................................................125
Figure B-70: South Cobb Dr. & I-285 SB On/Off Ramps, Video Capture, Location 1 04/04/2016 (Before). .......................................................................................125
Figure B-71: South Cobb Dr. & I-285 SB On/Off Ramps, Video Capture, Location 2 04/04/2016 (Before). .......................................................................................126
Figure B-72: Clairmont Rd. & I-85 SB, near Sam's Club, Source: Google Street View. ..................................................................................................................126
Figure B-73: Clairmont Rd. & I-85 SB, near Sam's Club, Video Capture, Location 1 04/12/2016 (Before). .......................................................................................127
Figure B-74: Clairmont Rd. & I-85 SB, near Sam's Club, Video Capture, Location 2 04/12/2016 (Before). .......................................................................................127
Figure B-75: Clairmont Rd. & I-85 SB, near Sam's Club, Location 1 04/04/2017 (After). ................................................................................................................128
87

Operational Evaluation of Do Not Block the Box Campaigns in Georgia Figure B-76: Clairmont Rd. & I-85 SB, near Sam's Club, Video Capture, Location 2
04/04/2017 (After)...........................................................................................128 Figure B-77: Clairmont Rd. & I-85 SB, near Sam's Club, Before and After Lane
Configuration and Camera Location. .................................................................129 Figure B-78: 14th St. NW & Hemphill Ave. NW, Source: Google Street
View. ..................................................................................................................129 Figure B-79: 14th St. NW & Hemphill Ave. NW, Video Capture, Location 1
04/14/2016 (Before). ..........................................................................................130 Figure B-80: 14th St. NW & Hemphill Ave. NW, Video Capture, Location 2
04/14/2016 (Before). ..........................................................................................130 Figure B-81: 14th St. NW & Hemphill Ave. NW, Video Capture, Location 1
05/01/2017 (After)..............................................................................................131 Figure B-82: 14th St. NW & Hemphill Ave. NW, Video Capture, Location 2
05/01/2017 (After)..............................................................................................131
88

Operational Evaluation of Do Not Block the Box Campaigns in Georgia

Table B-1: List of Intersections.

No. Intersection Name

Before

Intersection

After

After

"DBTB" Selection Result "DBTB"

"DBTB"

Data

Data

Data

Collection

Collection

Processing

Status

Status

Status

1 Parkside Pl. & Perimeter

Insufficient

Not selected

NA

Summit Pkwy.

capacity impact for after study

observed (merging

cases)

2 Ashford Dunwoody Rd. NE

Selected for after



& Ravinia Dr. NE

study

3 Peachtree Dunwoody Rd. &

Selected for after



Lake Hearn Dr. NE

study

4 Peachtree Dunwoody Rd. &

Selected for after



Johnson Ferry Rd.

study

5 Peachtree Dunwoody Rd. &

Selected for after



Abernathy Rd. NE

study

6 Peachtree Rd. NE & Lenox

Insufficient

Not selected

NA

Rd. NE

capacity impact for after study

observed

7 Peachtree Rd. NE &

Insufficient

Not selected

NA

Highland Dr. NE

capacity impact for after study

observed

8 Peachtree Rd. NE &

Selected for after (Only one

NA

Stratford Rd. NE

study

camera angle

recording)

9 Ponce De Leon Ave. NE &

Insufficient

Not selected

NA

City Hall East

capacity impact for after study

observed

10 S. Cobb Dr. & Pearl St.

Insufficient

Not selected

NA

capacity impact for after study

observed

11 S. Cobb Dr. & Walker St.

Selected for after Not painted

NA

study

12 W. Paces Ferry Rd. NW & I-

Selected for after



75 SB On/Off Ramps

study

13 Mount Paran Rd. NW at I-75

Insufficient

Not selected

NA

NB Off Ramp

capacity impact for after study

observed

14 Williams St. NW & 10th St.

Selected for after

Not

NW

study

Complete

15 South Cobb Dr. & I-285 SB

Selected for after Not painted

NA

On/Off Ramps

study

16 Clairmont Rd. & I-85 SB,

Selected for after



near Sam's Club

study

17 14th St. NW & Hemphill

Selected for after

Not

Ave. NW

study

Complete

89

Operational Evaluation of Do Not Block the Box Campaigns in Georgia

Table B-2: Hours of Video Data Collected and Processed.

No. Intersection Name

Days Video Data
Recorded
Before After

Total Hours of Data
Recorded
Before After

Total Hours of Data Initial Reviewed
Before After

1 Parkside Pl. &

9

0

67.5 NA 67.5 NA

Perimeter Summit

Pkwy.

2 Ashford Dunwoody 4

6

30

68.75 30

30

Rd. NE & Ravinia Dr.

NE

3 Peachtree Dunwoody 9

7

67.5 82.25 67.5 30

Rd. & Lake Hearn Dr.

NE

4 Peachtree Dunwoody 4

5

30

65.25 30

25

Rd. & Johnson Ferry

Rd.

5 Peachtree Dunwoody 4

4

30

56

30

20

Rd. & Abernathy Rd.

NE

6 Peachtree Rd. NE & 6

0

70.5 NA 28

NA

Lenox Rd. NE

7 Peachtree Rd. NE & 6

0

71.5 NA 28

NA

Highland Dr. NE

8 Peachtree Rd. NE & 6

8

70.5 92.25 28

30

Stratford Rd. NE

9 Ponce De Leon Ave. 5

0

56

NA 23

NA

at City Hall East

10 S. Cobb Dr. & Pearl 6

0

68.25 NA 28

NA

St.

11 S. Cobb Dr. & Walker 6

0

69

NA 28

NA

St.

12 W. Paces Ferry Rd. 6

9

67.5 92.5 25

30

NW & I-75 SB

On/Off Ramps

13 Mount Paran Rd. NW 6

0

68

NA 28

NA

at I-75 NB Off Ramp

14 Williams St. NW & 5

11

56.75 159.5 25

40

10th St. NW

15 South Cobb Dr. & I- 6

0

61.25 NA 28

NA

285 SB On/Off

Ramps

16 Clairmont Rd. & I-85 6

10

66

119.2 25

35

SB, near Sam's Club

5

17 14th St. NW &

6

10

71.25 138 25

35

Hemphill Ave. NW

NA Intersection not selected for after DBTB study. Video data not collected.

0 Intersection dropped from study. Video data not processed.

Total Hours of Data
Processed Before After
67.5 NA

12.5 9

18.5 6.5

11

8.75

13.5 4

0

NA

0

NA

0

NA

0

NA

0

NA

0

NA

6.75 9

0

NA

0

0

0

NA

10.5 9

0

0

90

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-1: Parkside Pl. & Perimeter Summit Pkwy., Source: Google Street View.
Figure B-2: Parkside Pl. & Perimeter Summit Pkwy., Video Capture, Location 1 10/23/2015 (Before). 91

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-3: Parkside Pl. & Perimeter Summit Pkwy., Video Capture, Location 2 10/23/2015 (Before).
Figure B-4: Ashford Dunwoody Rd. NE & Ravinia Dr. NE, Source: Google Street View. 92

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-5: Ashford Dunwoody Rd. NE & Ravinia Dr. NE, Video Capture, Location 1 10/26/2015 (Before).
Figure B-6: Ashford Dunwoody Rd. NE & Ravinia Dr. NE, Video Capture, Location 2 10/26/2015 (Before). 93

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-7: Ashford Dunwoody Rd. NE & Ravinia Dr. NE, Video Capture, Location 3 10/26/2015 (Before).
Figure B-8: Ashford Dunwoody Rd. NE & Ravinia Dr. NE, Video Capture, Location 4 10/26/2015 (Before). 94

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-9: Ashford Dunwoody Rd. NE & Ravinia Dr. NE, Video Capture, Location 1 05/06/2016 (After).
Figure B-10: Ashford Dunwoody Rd. NE & Ravinia Dr. NE, Video Capture, Location 2 04/19/2016 (After). 95

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-11: Ashford Dunwoody Rd. NE & Ravinia Dr. NE, Before DBTB Lane Configuration and Camera Location.
Figure B-12: Ashford Dunwoody Rd. NE & Ravinia Dr. NE, After DBTB Lane Configuration and Camera Location. 96

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-13: Peachtree Dunwoody Rd. & Lake Hearn Dr. NE, Source: Google Street View.
Figure B-14: Peachtree Dunwoody Rd. & Lake Hearn Dr. NE, Video Capture, Location 1 10/23/2015 (Before). 97

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-15: Peachtree Dunwoody Rd. & Lake Hearn Dr. NE, Video Capture, Location 2 10/23/2015 (Before).
Figure B-16: Peachtree Dunwoody Rd. & Lake Hearn Dr. NE, Video Capture, Location 3 10/23/2015 (Before). 98

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-17: Peachtree Dunwoody Rd. & Lake Hearn Dr. NE, Video Capture, Location 1 05/06/2016 (After).
Figure B-18: Peachtree Dunwoody Rd. & Lake Hearn Dr. NE, Video Capture, Location 2 05/06/2016 (After). 99

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-19: Peachtree Dunwoody Rd. & Lake Hearn Dr. NE, Before Lane Configuration and Camera Location.
Figure B-20: Peachtree Dunwoody Rd. & Lake Hearn Dr. NE, After Lane Configuration and Camera Location. 100

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-21: Peachtree Dunwoody Rd. & Johnson Ferry Rd., Source: Google Street View.
Figure B-22: Peachtree Dunwoody Rd. & Johnson Ferry Rd., Video Capture, Location 1 10/26/2015 (Before). 101

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-23: Peachtree Dunwoody Rd. & Johnson Ferry Rd., Video Capture, Location 2 10/26/2015 (Before).
Figure B-24: Peachtree Dunwoody Rd. & Johnson Ferry Rd., Video Capture, Location 3 10/26/2015 (Before). 102

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-25: Peachtree Dunwoody Rd. & Johnson Ferry Rd., Video Capture, Location 4 10/26/2015 (Before).
Figure B-26: Peachtree Dunwoody Rd. & Johnson Ferry Rd., Video Capture, Location 1 04/25/2016 (After). 103

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-27: Peachtree Dunwoody Rd. & Johnson Ferry Rd., Video Capture, Location 2 04/25/2016 (After).
Figure B-28: Peachtree Dunwoody Rd. & Johnson Ferry Rd., Before Lane Configuration and Camera Location. 104

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-29: Peachtree Dunwoody Rd. & Johnson Ferry Rd., After Lane Configuration and Camera Location.
Figure B-30: Peachtree Dunwoody Rd. & Abernathy Rd. NE, Source: Google Street View. 105

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-31: Peachtree Dunwoody Rd. & Abernathy Rd. NE, Video Capture, Location 1 10/28/2015 (Before).
Figure B-32: Peachtree Dunwoody Rd. & Abernathy Rd. NE, Video Capture, Location 2 10/28/2015 (Before). 106

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-33: Peachtree Dunwoody Rd. & Abernathy Rd. NE, Video Capture, Location 3 10/28/2015 (Before).
Figure B-34: Peachtree Dunwoody Rd. & Abernathy Rd. NE, Video Capture, Location 4 10/28/2015 (Before). 107

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-35: Peachtree Dunwoody Rd. & Abernathy Rd. NE, Video Capture, Location 1 04/18/2016 (After).
Figure B-36: Peachtree Dunwoody Rd. & Abernathy Rd. NE, Video Capture, Location 2 04/18/2016 (After). 108

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-37: Peachtree Dunwoody Rd. & Abernathy Rd. NE, Before Lane Configuration and Camera Location.
Figure B-38: Peachtree Dunwoody Rd. & Abernathy Rd. NE, After Lane Configuration and Camera Location. 109

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-39: Peachtree Rd. NE & Lenox Rd. NE, Source: Google Street View.
Figure B-40: Peachtree Rd. NE & Lenox Rd. NE, Video Capture, Location 1 05/02/2016 (Before).
110

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-41: Peachtree Rd. NE & Highland Dr. NE, Source: Google Street View.
Figure B-42: Peachtree Rd. NE & Highland Dr. NE, Video Capture, Location 1 05/02/2016 (Before). 111

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-43: Peachtree Rd. NE & Stratford Rd. NE, Source: Google Street View.
Figure B-44: Peachtree Rd. NE & Stratford Rd. NE, Video Capture, Location 1 05/02/2016 (Before). 112

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-45: Peachtree Rd. NE & Stratford Rd. NE, Video Capture, Location 1 04/19/2017 (After).
Figure B-46: Peachtree Rd. NE & Stratford Rd. NE, Video Capture, Location 2 04/19/2017 (After). 113

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-47: Ponce De Leon Ave. at City Hall East, Source: Google Street View.
Figure B-48: Ponce De Leon Ave. at City Hall East, Video Capture, Location 1 03/18/2016 (Before).
114

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-49: S. Cobb Dr. & Pearl St., Source: Google Street View.
Figure B-50: S. Cobb Dr. & Pearl St., Video Capture, Location 1 03/21/2016 (Before).
115

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-51: S. Cobb Dr. & Pearl St., Video Capture, Location 2 03/21/2016 (Before).
Figure B-52: S. Cobb Dr. & Walker St., Source: Google Street View. 116

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-53: S. Cobb Dr. & Walker St., Video Capture, Location 1 03/21/2016 (Before).
Figure B-54: S. Cobb Dr. & Walker St., Video Capture, Location 2 03/21/2016 (Before).
117

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-55: W. Paces Ferry Rd. NW & I-75 SB On/Off Ramps, Source: Google Street View.
Figure B-56: W. Paces Ferry Rd. NW & I-75 SB On/Off Ramps, Video Capture, Location 1 03/28/2016 (Before). 118

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-57: W. Paces Ferry Rd. NW & I-75 SB On/Off Ramps, Video Capture, Location 2 03/28/2016 (Before).
Figure B-58: W. Paces Ferry Rd. NW & I-75 SB On/Off Ramps, Video Capture, Location 1 04/04/2017 (After). 119

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-59: W. Paces Ferry Rd. NW & I-75 SB On/Off Ramps, Video Capture, Location 2 04/04/2017 (After).
Figure B-60: W. Paces Ferry Rd. NW & I-75 SB On/Off Ramps, Before and After Lane Configuration and Camera Location. 120

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-61: Mount Paran Rd. NW at I-75 NB Off Ramp, Source: Google Street View.
Figure B-62: Mount Paran Rd. NW at I-75 NB Off Ramp, Video Capture, Location 1 03/28/2016 (Before). 121

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-63: Mount Paran Rd. NW at I-75 NB Off Ramp, Video Capture, Location 2 03/28/2016 (Before).
Figure B-64: Williams St. NW & 10th St. NW, Source: Google Street View. 122

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-65: Williams St. NW & 10th St. NW, Video Capture, Location 1 04/06/2016 (Before).
Figure B-66: Williams St. NW & 10th St. NW, Video Capture, Location 2 04/06/2016 (Before).
123

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-67: Williams St. NW & 10th St. NW, Video Capture, Location 1 05/09/2017 (After).
Figure B-68: Williams St. NW & 10th St. NW, Video Capture, Location 2 05/09/2017 (After).
124

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-69: South Cobb Dr. & I-285 SB On/Off Ramps, Source: Google Street View.
Figure B-70: South Cobb Dr. & I-285 SB On/Off Ramps, Video Capture, Location 1 04/04/2016 (Before). 125

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-71: South Cobb Dr. & I-285 SB On/Off Ramps, Video Capture, Location 2 04/04/2016 (Before).
Figure B-72: Clairmont Rd. & I-85 SB, near Sam's Club, Source: Google Street View. 126

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-73: Clairmont Rd. & I-85 SB, near Sam's Club, Video Capture, Location 1 04/12/2016 (Before).
Figure B-74: Clairmont Rd. & I-85 SB, near Sam's Club, Video Capture, Location 2 04/12/2016 (Before). 127

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-75: Clairmont Rd. & I-85 SB, near Sam's Club, Location 1 04/04/2017 (After).
Figure B-76: Clairmont Rd. & I-85 SB, near Sam's Club, Video Capture, Location 2 04/04/2017 (After). 128

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-77: Clairmont Rd. & I-85 SB, near Sam's Club, Before and After Lane Configuration and Camera Location.
Figure B-78: 14th St. NW & Hemphill Ave. NW, Source: Google Street View. 129

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-79: 14th St. NW & Hemphill Ave. NW, Video Capture, Location 1 04/14/2016 (Before).
Figure B-80: 14th St. NW & Hemphill Ave. NW, Video Capture, Location 2 04/14/2016 (Before).
130

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure B-81: 14th St. NW & Hemphill Ave. NW, Video Capture, Location 1 05/01/2017 (After).
Figure B-82: 14th St. NW & Hemphill Ave. NW, Video Capture, Location 2 05/01/2017 (After).
131

Operational Evaluation of Do Not Block the Box Campaigns in Georgia 132

Appendix C

VIDEO TOOL FOR MANUALLY EXTRACTING COMPLEX TRAFFIC DATA

LIST OF FIGURES
Figure C-1: Snapshot of the Three Windows of GT-MVP's Blocking Event Extraction Module. .............................................................................................139
Figure C-2: GT-MVP's Video Player Window Interface.......................................140 Figure C-3: GT-MVP's Blocking Event Extraction Interface Window, Customized
for Do-Not-Block-the-Box Study.......................................................................141 Figure C-4: Structural Frame of GT-MVP's GUI. .................................................142 Figure C-5: Control Flow of GT-MVP GUI...........................................................143 Figure C-6: GT-MVP Being Used for Blocking Event Data Extraction. ...............147 Figure C-7: GT-MVP's Propensity Extraction Interface Window. ........................149
C.1 Abstract
Driver behavior studies often require the analysis of highly specific and customized observational data attributes. Due to the complex nature of behavioral data collection, typically requiring some degree of customization depending on the study, there is a lack of generic tools available to transportation professionals to assist in the data collection process. This paper presents a python-based software application "GT-MVP" designed to provide a user-friendly interface to manually collect complex video-based traffic data. GTMVP's graphical user interface allows users to play multiple videos and operate them synchronously using common controls and to easily review and correct errors during the data collection session. GT-MVP has been used to improve the efficiency of manual data collection in a study of vehicle blocking behavior at intersections. Compared to previous
133

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
approaches used to collect behavioral data required for this study, GT-MVP took 65% less time and reduced the missed detection rate. GT-MVP interface can be modified to assist in the collection of complex traffic data for different traffic studies and can also be used to improve efficiency of collection of basic traffic data such as vehicle counts and is available to the community as an open source software.
Keywords: Video data extraction, Data collection, Video annotation, Don't Block the Box
C.2 Introduction
Advancements in traffic data collection technologies have steadily increased the amount of data available to transportation researchers and practitioners. Fixed sensorbased technologies, including intrusive technologies such as inductive loops, pneumatic road tubes, piezoelectric sensors, etc. and non-intrusive technologies such as microwave, RADAR, and video detection systems generate screen-line counts, point speed data, etc. Advancements in using wireless technologies such as GPS, mobile phones, Bluetooth, and on-road sensors, etc. have enabled probe-based data collection with improved traveltime and space-mean speed data.
In spite of advancements in the efficiency of automatic sensorbased technologies, the roadway environment can adversely affect the accuracy of data generated by these systems [1]. These impacts, when coupled with significant deployment costs can limit the feasibility of using these systems for short-term data collection needs. Traditionally, where permanent technologies are not available or feasible, temporary devices such as video recorders are utilized, with data manually extracted from the video. However, such manual data extraction can be time consuming and costly, motivating interest in automated video-
134

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
based data collection systems to extract data [2]. Video analysis tools for trajectory determination, vehicle speed measurement, vehicle counts, and vehicle classification, etc. have been developed to study roadway safety and operation [27] as well as for vehicle and pedestrian behavioral studies [4, 8]. While these tools have been used for decisionbased data extraction without human input for a number of traffic, pedestrian, and behavioral studies [4, 6, 7, 9, 10], many studies may have complex and subjective data extraction requirements that are difficult to fully automate. Existing automated data collection tools may be either insufficient or require considerable time for development of algorithms that can assess complex roadway and traffic conditions. For these applications, effective semi-automatic annotation tools that assist users in more effective manual data extraction are still needed [11].
To address the need for a customizable and efficient interface to aid in manually extracting high resolution and complex data to understand driver behavior, a video data extraction tool, Georgia Tech's - Multi Video Player (GT-MVP), has been developed. This tool allows a user to manually generate data for studies that require basic data (such as vehicle count, vehicle headways, signal timing, etc.) as well as for operational and behavioral studies that require complex data (such as classification of vehicles or pedestrians based on behavior).
The paper is organized as follows: the background section contains studies related to video annotation tools. Development and user interface design of the GT-MVP tool are included under the "Multi-video Player Application" section. Several case studies are presented to demonstrate the use of the tool. Benefits of using GT-MVP are quantified in
135

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
the "Verification of GT-MVP" section. Finally, the application, benefits, limitations, and potential future work are summarized in the "Conclusions" section.
C.3 Background
Driver behavior studies often require the analysis of highly specific and customized observational data attributes. As such, data collection/extraction protocols typically require some degree of study-specific customization and there is a lack of generic tools available to transportation professionals to assist in the data collection process. In the field of computer vision, a class of tools, known as video annotation tools, provides conceptually similar functionality to the requirements of driver behavior studies, albeit for a different application case. The development and design aspects of video annotation tools are relevant to development of the GT-MVP tool. The significant difference is that video annotation tools focus on the functionality of being able to render overlaid bounding boxes on top of the video, while the GT-MVP tool focuses on data attributes recorded by making manual observations in the frame of the video.
Video annotation has long been a critical step in the research and development process in computer vision. Researchers use video-annotated data to develop algorithms that may be applied in an automated video data extraction system to understand events, recognize objects, and predict future events. Example interactive video annotation tools include LabelMe [12], ViPER [13], GTVT [9], VATIC [14], and GTGT [10]. LabelMe allows a user to annotate objects/humans of different shapes in a video. This online, webbased platform allows for extraction of complex event annotations from high-quality videos. The annotation process requires a user to draw a polygon by clicking around the object boundary. The user navigates through the video using video controls and edits the
136

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
polygon position as the object location changes [11, 12, 14]. Mihalcik et al. presented a flexible video annotation tool ViPER (Video Performance Evaluation Resource)[13]. ViPER allows for frame-by-frame annotation. File information such as content description, date, and keywords associated with content can be added. It can be used for many purposes such as tracking people, detecting text, etc. [11, 13, 16]. Data obtained using ViPER have been utilized to evaluate computer vision algorithms for text detection, face detection, and vehicle detection, etc. [15, 17]. In 2009, Ambardekar et al. proposed GTVT (Ground Truth Verification Tool) for video surveillance systems. GTVT focuses on object detection and classification [9]. In 2014, Mossi et al. proposed a ViPER set up with a simple visual interface and a jog shuttle wheel to speed up generation of ground truth data of traffic systems. This study also asserted that for generating ground-truth data to measure basic traffic system measurements there is no need of pixel level segmentation or frame-by-frame tracking [15]. In 2012, another video annotation platform, VATIC (Video Annotation Tool from Irvin, California) was introduced. This tool is designed to assist computer vision research. It uses Amazon's "Mechanical Turk" for crowdsourcing video annotation. Some extensions developed for VATIC include tracking integration with OpenCV, sentence annotations, time interval labeling, and human action labeling, etc. [14, 18]. Designed to generate databases for computer vision research, the interface is rather complicated for data collection of simple traffic measurements. In 2016, Bigaj et al. designed an application for video detection purposes that is targeted toward traffic system video analysis, focusing on a data generation system with just a single mouse click, GTGT (Ground Truth Generation Tool) simplified the extraction of road traffic data [10].
137

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Studies requiring vehicle identification across multiple video time steps generally require vehicle re-identification or classification. Re-identification is significantly more complex than presence detection. The current study develops GT-MVP to specifically address manual data collection where re-identification is required, using a semi-automatic procedure that provides the users with an efficient interface to extract low-level traffic data such as vehicle arrivals and departures at specific screen-lines and vehicle behaviorbased classifications. Based on insights gained from previous efforts and the literature, GT-MVP seeks to address two challenges: 1) develop a simple interface that may be quickly adapted by new users, and 2) address data collection where multiple views are necessary to extract the needed information. While the first goal is intended to improve efficiency and accuracy in data collection, the second addresses those studies where a single view is often insufficient. Thus, GT-MVP offers a unified interface to synchronize and play multiple videos, thus providing a comprehensive field of view. GT-MVP also offers controls to play multiple videos faster or slower than recorded time or frame-by-frame (backward and forward) in synchronized mode for data extraction. GT-MVP also improves data extraction efficiency by providing an optimized user-friendly graphical user interface (GUI) that minimizes human effort and allows for review and correction of errors.
C.4 Multi-video Player Application
The GT-MVP application has been developed primarily to address manual traffic-related data extraction from pre-recorded videos. GT-MVP is utilized to support transportation research studies that require complex vehicle/driver behavior information in addition to regular data attributes such as vehicle volume counts, signal phase information, etc. A brief
138

Operational Evaluation of Do Not Block the Box Campaigns in Georgia description of the GT-MVP application is provided in this section, followed by some case studies that demonstrate specific tool applications.
GT-MVP is developed on the Python platform. The user interface consists of two video player windows and a control window with common video player controls to operate both video player windows simultaneously (see Figure C-1). The common control window, also referred as the data extraction window, provides an interface to assist the user to record the required data in a semi-automatic fashion, minimizing typing by leveraging the framenumbers and embedded timestamps of the video stream. Figure C-2: provides a close-up of a single GT-MVP video player window interface.
Figure C-1: Snapshot of the Three Windows of GT-MVP's Blocking Event Extraction Module.
Apart from a few generic data collection interfaces, the widgets on the data extraction window are amenable to customization to the requirements of specific studies. For example, it will be seen later that the data extraction window graphical user interface had
139

Operational Evaluation of Do Not Block the Box Campaigns in Georgia been customized for data collection for an ongoing "Do-Not-Block-the-Box" study. The Event Extraction Interface window (see Figure C-3: ) contains the controls required to play the multiple videos concurrently, as well as the user interface to extract and store blocking event data.
Figure C-2: GT-MVP's Video Player Window Interface.
140

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure C-3: GT-MVP's Blocking Event Extraction Interface Window, Customized for Do-Not-Block-the-Box Study.
Figure C-4: shows GT-MVP's structural framework. As seen, GT-MVP's graphical user interface is developed using several widgets under Python's Tkinter library [19]. Each video player window contains three Tkinter frames: one frame displays an embedded VLC media player instance, another frame contains the video time slider, and the third frame contains player controls as button widgets and Tkinter entry cells [1921]. The data extraction window contains two Tkinter canvases. The first canvas contains a frame which holds Tkinter buttons and entry cells for common controls, data extraction buttons (for example, to retrieve the video timestamp), lane movement and vehicle types entry cells, etc. The second canvas contains a frame that consists of a matrix of entry cells. Horizontal and vertical scrollbars are embedded in the second canvas to ease navigation of this matrix. The frame and widgets in the canvas are not affected by vertical and horizontal scrollbars
141

Operational Evaluation of Do Not Block the Box Campaigns in Georgia present in canvas 2, thus keeping frame 1 frozen while allowing a user to scroll down the cell matrix where data are stored.
Figure C-5: shows the control flow of the GT-MVP graphical user interface. As seen, the graphical user interface of GT-MVP contains two units of control flow--the video player loop and the data extraction loop. The video player loop contains player controls to operate videos individually as well as common player controls to operate all videos together. When the user interacts with individual player controls respective videos are rendered and when the user interacts with common controls all videos are rendered. The data extraction loop controls user interactions with GT-MVP to extract data from the video. Input is received from the user in the form of mouse click or keyboard press events and extracted data are saved to a CSV file. For instance, when the user clicks the "Get Timestamp" button on the data extraction window, the time instant of last rendered video instance in the video player loop is stored in the data extraction window. If the user stops data collection in the middle of a video, data from the CSV file can be loaded back to the interface to continue data extraction at a later time.
Figure C-4: Structural Frame of GT-MVP's GUI. 142

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure C-5: Control Flow of GT-MVP GUI. C.4.1 Case Study: Data Extraction for Intersection Blocking Study The primary motivation for the development of GT-MVP was the data extraction requirements for a study exploring the impacts of vehicle-blocking behavior on traffic operations. In this study, a vehicle is considered to "Block the Box" when it enters the intersection with no space to exit due to the traffic spillback downstream. Use of a "Don't Block the Box" (DBTB) treatment as a control strategy has been adopted by many cities in the U.S. and this research explores the performance of DBTB treatments by quantifying the impact of vehicle blocking behavior on intersection performance [22].
DBTB data analysis includes comparing the characteristics of blocking events and measuring the propensity to block for vehicles on a given intersection approach (i.e., the likelihood of a vehicle to enter an intersection when it cannot exit due to insufficient space downstream of intersection) before and after DBTB signing and marking implementation. To determine the effectiveness of DBTB treatments, blocking event characteristics such as number of blocking events, event duration, vehicle lane movements that block or are blocked, etc., are compared before and after DBTB treatment implementation. In addition, the propensity to block is compared to study the impact of DBTB treatments on driver
143

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
behavior as well as provide a calibration parameter to allow for simulation of queuing, delay, and travel time analysis with and without a DBTB treatment [22].
To study the blocking event characteristics and to measure a vehicle's propensity to block, two modules of GT-MVP application were developed: 1) a Blocking Event Data Extraction GUI, and 2) a Propensity to Block Data Extraction GUI. In both GUIs, each video player window has a time slider below to indicate the playback position in the video track (see Figure C-1). Video player control buttons below the time slider allow the user finer control over playback. In addition to basic player operations such as Play and Pause, the "Play (0.5X)" and "Play (2X)" buttons enable video playback at half of original speed or twice the original speed, allowing the user to move through the video more efficiently. Backward skip and Forward skip buttons skip the video frame back and forth respectively by the number of seconds the user provides in the "Custom Skip" entry field, providing further flexibility in traversing the video timeline. (For example, in the DBTB efforts a custom skip on the order of half the cycle length of the given intersection allowed efficient skipping of video not relevant to the data collection.) Further details of the features of the data recording window are provided in the case studies. The next subsections provide a summary of data required for the DBTB study and a comparison of the traditional method to extract data from video versus using GT-MVP application modules.
C.4.2 Blocking Event Data Extraction A blocking event is defined to occur at the start of the green indication of the conflicting phase where the blocking vehicle was not able to exit the intersection right-of-way during their phase. The blocking event ends when the blocking vehicle departs the intersection right-of-way or "box." An approach lane that is blocked is referred to as the "Blocked
144

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Lane" and the approach lane from which the blocking vehicle entered the intersection is referred to as the "Block-Source Lane."
Before the development of the GT-MVP application, data for each blocking event was extracted by playing back the videos in a media player and recording the observations in a spreadsheet manually. A challenge in this data collection was that a different video angle was consistently needed to capture the signal phase data than the vehicles in the blocking event. Thus, the researcher conducting the data extraction was required to manage videos in separate players. The data attributes manually recorded in the spreadsheet included a unique serial ID of the blocking event, the start time of blocking, the end time of the blocking event, and the block source and blocked lanes (coded as a three-digit number). Users recorded the lanes associated with blocking event by entering "1" in the cell below the corresponding lane movement code. The duration of blocking was derived as the difference of the blocking event start and end time.
In a second pass through the videos the blocking event was inspected to determine if capacity was affected. Data were collected regarding whether the blocking event obstructed any vehicles with the right-of-way, whether there was a presence of demand during the blocking event, and whether the event was a full or partial blocking based on whether the blocked vehicles were unable to proceed or were hindered but able to complete their movement. Other relevant details about the blocking event such as impacts on safety of pedestrian crossing or ambiguities in any recorded data were documented under the comments column.
145

Operational Evaluation of Do Not Block the Box Campaigns in Georgia C.4.3 Using GT-MVP's Blocking Event Extraction Module for Data Extraction
In GT-MVP the two intersection videos are played in a synchronized manner using GTMVP's common controls features. Any offset between two videos is eliminated by going backward or forward using the custom skip functionality in one of the video player windows. After the initial synchronization, the common controls in the data extraction window are used to play (play, pause, play at slow rate, play at faster rate, skip forward or backward, etc.) the two videos concurrently. By clicking relevant widgets on the data extraction interface, blocking event details data are automatically and instantly recorded in the empty cell matrix in the common window. The empty cell matrix consists of columns to store all required blocking data. Mouse clicks on "Get Timestamp," "Lane movement number code," "Yes" or "No" buttons under vehicle obstruction, presence of demand, and capacity impact label, records data in the first empty cell under the corresponding column. In case of a user error during the data extraction process, the recorded data in the cell can be directly edited as with a spreadsheet, providing a familiar interface to the user. Figure C-6: shows a snapshot of GT-MVP being used for blocking event data extraction.
146

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure C-6: GT-MVP Being Used for Blocking Event Data Extraction. C.4.4 Propensity to Block Data Extraction In the DBTB study analysis, estimation of the drivers' propensity to block was crucial to modeling the impact of blocking on traffic flow. Drivers' propensity to block is measured on the "Block-Source Lane," i.e. the lane from which the blocking vehicles originate. In this effort, a drivers' propensity to block for an intersection-lane is defined as the likelihood of a driver to enter the intersection and block when there is no space to exit downstream of the intersection (i.e., when presented with a blocking opportunity or choice). Drivers' propensity to block is field measured as the ratio of number of vehicles that enter and block the intersection to the total number of vehicles that have an opportunity to block.
Prior to the development of the GT-MVP tool, propensity data were extracted using another in-house developed video player software called "Videoanalyzer"[23]. Videoanalyzer allowed for recording vehicle timestamps from the block source lane when they crossed the stop bar. However, the timestamp recorded by Videoanalyzer was the
147

Operational Evaluation of Do Not Block the Box Campaigns in Georgia computer's clock time, rather than the video stream time. This added some complexities to the post processing of the data as the timestamps had to be adjusted after the data extraction process, to coincide with the video time. In addition, Videoanalyzer did not have the capability of displaying the recorded data for review during the data extraction process, which limited the ability to recognize and correct errors. A second pass through the video was done to tag the vehicle IDs in the resulting data as blocking, non-blocking or notapplicable (i.e., if space existed to exit the intersection, the driver does not have an opportunity to block). The need for a second pass significantly increased the data reduction time. In addition, the lack of options to forward and reverse the video in Videoanalyzer made revisiting and re-recording data time consuming. These drawbacks were addressed in the GT-MVP application.
C.4.5 Using GT-MVP's Propensity Extraction Module for Data Extraction The Data Extraction Window of GT-MVP's Propensity Extraction Module includes common controls to operate the two video player windows simultaneously and also includes buttons and an integrated spreadsheet to extract propensity data from the videos, as shown in Figure C-7: .
148

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Figure C-7: GT-MVP's Propensity Extraction Interface Window. Videos captured from two camera angles are synchronized in the two video player windows. Vehicles on the "block source" lane are observed. If the vehicle has no space to exit the intersection then the timestamp when it crosses the stop bar is recorded with a mouse click on "Get Timestamp." Based on the vehicle's action to enter or not enter the intersection the Blocking or Non-Blocking button is clicked to record the vehicle behavior. Thus, propensity data are extracted in a single review of the video. GT-MVP also allows a user to add comments where necessary. The data extracted are displayed in the first empty cell under corresponding column, and again may be edited should an error occur. C.4.6 Other Potential Application Cases Apart from the case study described above, there is a wide variety of other studies where GT-MVP may provide efficiency improvements in the video data extraction process.
149

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
Several studies the research team has undertaken over the past several years provide excellent examples, such as the following:
Roundabout Capacity Calibration Study. For a traffic study conducted to calibrate the HCM 2010 roundabout capacity equations to include characteristics of Georgia drivers, gap acceptance field data were extracted from video recordings. The study included an investigation of the impact on capacity due to the hesitation of entering vehicles to enter the roundabout when faced with the uncertainty related to whether vehicles in the roundabout are exiting or circulating. The study required the simultaneously recording of two legs of a roundabout. Timestamps for circulating, entering, and exiting vehicles were recorded requiring multiple video viewing. [24]. GT-MVP would have provided an integrated GUI for data-collectors to process both videos simultaneously and record the relevant interface with a single concurrent traversal of the videos.
Left-Turn Interactions Safety Study. This traffic study evaluated effectiveness of post encroachment time (PET) and acceleration-deceleration profiles of vehicles as surrogates for safe left-turn interactions. The PET and acceleration-deceleration data were used to develop a crash data model. For this study the vehicle trajectory over a stretch of road (high speed intersection approach) was tracked to obtain the acceleration-deceleration profiles. Due to the limitation of the field of view of a single camera, two cameras were used to record a sufficient length of roadway. As video synchronization was critical for this effort, a custom software was developed to synchronize videos; however, the interface lacked a run-time review of the data capability, limiting the ability of the user to quality
150

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
check the data during the data collection process [25]. This limitation has been addressed in GT-MVP through a spreadsheet interface present on the data extraction window.
Vehicle and Pedestrian Behavior under Varying Crosswalk Treatments. Recently, members of the research team investigated the effectiveness of various crosswalk treatments to elicit yielding behavior from crossing vehicles. It was seen that two camera angles were typically necessary to allow for the inference of the crossing intent of the pedestrian while tracking the position of approaching vehicles. The ability to readily synchronize the videos as well as flexibility in forwarding and reversing through the video in time steps would have significantly improved the efficiency of the data reduction team.
C.5 Verification of GT-MVP: Propensity to Block Data Extraction
To quantify the benefits of extracting data using the GT-MVP tool versus using the video playback with manual data entry methods, an experiment was performed where the previously described blocking propensity data were extracted from the same videos using Videoanalyzer and GT-MVP. Different data collectors were assigned to perform the extraction process from the two methods to avoid any bias related to previous knowledge about the events in the videos affecting the data extraction process. While there are some possible differences between efficiency of the data collectors, they all had experience from previous projects and differences are expected to be minimal. An hour of video at a single intersection was used for analysis. These data had been previously analyzed with the results visually inspected for accuracy. Comparisons were performed based on the total time
151

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
required for extraction, errors in capturing events (missing data), and misclassification of data (errors in data).
Out of 160 vehicle timestamps identified in GT-MVP, seven were missed by the data collector using Videoanalyzer. Videoanalyzer's failure to allow a simple procedure for error correction and easily traverse the video is hypothesized to deter users from going back to revisit and confirm any suspected errors during the data extraction. In addition, 3 out of 36 blocking vehicles were misclassified by Videoanalyzer users, which is likely at least partially attributable to the tedious nature of the extraction process involved and the need for multiple passes through the video. For the overall data extraction process, data collectors required 100 minutes and 35 minutes to process 1 hour of video using Videoanalyzer and GT-MVP, respectively representing a nearly two-thirds reduction in processing time.
C.6 Conclusions
The GT-MVP system was developed to allow users an effective semi-automated procedure for extracting a variety of important data from video recordings to support a wide range of traffic and behavioral studies. The advanced playback features and the integrated data collection and review interface of GT-MVP seek to reduce the manual data extraction time and enhance data accuracy for extraction of information supporting these types of studies. In addition, the ability to play multiple videos concurrently effectively increases the width of field of view for the data collector. For example, for the blocking event data collection, the data collector was able to simultaneously observe the existence of demand, queues, and the signal heads while collecting timestamps, in a single pass through the videos. Compared to previous data extraction methods, e.g., earlier roundabout capacity calibration studies,
152

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
this feature would have significantly reduced the time required for data extraction by providing a single wide field of view. In addition, the wider field of view helps address issues related to occlusions from large vehicles common under normal field conditions.
The single unified control for simultaneous playback, rewind, and replay of multiple videos in a synchronous fashion increases the efficiency for traversing the videos as well as reduces the possibility of errors caused by switching between videos. The custom-skip feature allows the users to skip over periods with no events relevant to the study while allowing them to slow down the traversal during relevant events thus minimizing the chances of missing any important data.
The interface with preconfigured buttons for the relevant information capture reduces data errors related to typing errors while also ensuring that relevant information is captured with a reduced probability of error. For example, in the blocking event data extraction application, the automatic recording of timestamps initiated by a button click reduces the chances of error in comparison of timestamps entered manually in different cells in the spreadsheet.
The ability to review and edit the recorded data in run time allows the user to deal with erroneous data entries without requiring the user to manually edit the background data file as well as reducing potential errors. Similarly, automatically saving data into automatically named comma-separated-value (CSV) files simplifies subsequent analysis by producing data files that can be easily opened in a standard spreadsheet application or loaded directly into standard statistical software. In addition, this automatic naming of files eases the management and tracking of the datasets over the course of the project by ensuring that the location name, initials of data collector, and time of creation of file are
153

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
referenced to the data, over the period of the project. Similarly, auto-saving of data after each new entry or edit minimizes data loss and rework. The ability to reload data from an existing data file not only allows for continuity of work flow across data collection sessions but also provides the ability to subsequently review the collected data, which simplifies data quality control.
Although the GT-MVP application was developed to meet the needs of the DBTB study, it is applicable to a wide range of other traffic studies. The key feature of operating two video player windows simultaneously assists the user in collecting data with ease and more accuracy in a single integrated data file. The interface can be easily leveraged for other applications such as performing volume counts, turn movement counts at intersections, etc. GT-MVP can also be used for vehicle headway extraction for studies involving saturation flow rate determination.
This tool, developed on the Python platform, provides the flexibility of easy modifications by other researchers to modify the interface to record other data attributes specific to the study. One of the features that has been observed in other such studies is the ability to draw guide-lines or overlay grids on the video player pane. This is a feature planned for the next version of the tool. The source code of the application is available for use by the community at http://transportation.ce.gatech.edu/publications.
C.7 Acknowledgements
The information, data, or work presented herein was funded in part by the Georgia Department of Transportation as Research Project 14-43. The views and opinions of authors expressed herein do not necessarily state or reflect those of the State of Georgia or any agency thereof. This paper does not constitute a standard, specification, or regulation.
154

Operational Evaluation of Do Not Block the Box Campaigns in Georgia

C.8 References

1.

Leduc, G., Road Traffic Data: Collection Methods and Applications, European

Commission Joint Research Centre. Institute for Prospective Technological Studies,

Editor. 2008.

2.

Toth, C., et al., "Tablet-Based Traffic Counting Application Designed to Minimize Human

Error." Transportation Research Record: Journal of the Transportation Research Board,

2013. 2339: p. 3946.

3.

Jackson, S., et al., "Flexible, Mobile Video Camera System and Open Source Video

Analysis Software for Road Safety and Behavioral Analysis." Transportation Research

Record: Journal of the Transportation Research Board, 2013. 2365: p. 9098.

4.

Saunier, N., and T. Sayed, "Automated Analysis of Road Safety with Video Data."

Transportation Research Record: Journal of the Transportation Research Board, 2007.

2019: p. 5764.

5.

Strong, C., S. Lowry, and P. McCarthy, "Collecting Vehicle-Speed Data by Using Time-

Lapse Video Recording Equipment." Transportation Research Record: Journal of the

Transportation Research Board, 2003. 1855: p. 97104.

6.

Wu, J., and Y. Tsai, "Enhanced Roadway Geometry Data Collection Using an Effective

Video Log Image-Processing Algorithm." Transportation Research Record: Journal of the

Transportation Research Board, 2006. 1972: p. 133140.

7.

Zhang, G., R. Avery, and Y. Wang, "Video-Based Vehicle Detection and Classification

System for Real-Time Traffic Data Collection Using Uncalibrated Video Cameras."

Transportation Research Record: Journal of the Transportation Research Board, 2007.

1993: p. 138147.

8.

J. Blunsden, S., and R. Fisher, The BEHAVE video dataset: ground truthed video for multi-

person. Vol. 4., 2009.

155

Operational Evaluation of Do Not Block the Box Campaigns in Georgia

9.

Ambardekar, A., M. Nicolescu, and S. Dascalu. "Ground Truth Verification Tool (GTVT)

for Video Surveillance Systems." In 2009 Second International Conferences on Advances

in Computer-Human Interactions, 2009.

10. Szymon Bigaj, A.G., Jacek Kociow, Zbigniew Mikrut, Piotr Pawlik, "Ground Truth

Generating Tool for Traffic Video Detector." Image Processing & Communications, 2016. 20.2 p. 1727.

11. Biresaw, T.A., et al., "ViTBAT: Video tracking and behavior annotation tool." In 2016

13th IEEE International Conference on Advanced Video and Signal Based Surveillance (AVSS). 2016. 12. Jenny, Y., et al., "LabelMe video: Building a video database with human annotations." In

2009 IEEE 12th International Conference on Computer Vision. 2009. 13. Laboratory, L.a.M.P., ViPER: The Video Performance Evaluation Resource. [cited 2017
28 July]; Available from: http://viper-toolkit.sourceforge.net/.

14. Vondrick, C., D. Patterson, and D. Ramanan, Efficiently Scaling Up Crowdsourced Video

Annotation. International Journal of Computer Vision, 2013. 101(1): p. 184204.

15. Mossi, J.M., et al., Ground truth annotation of traffic video data. Multimedia Tools and

Applications, 2014. 70(1): p. 461474.

16. Doermann, D., and D. Mihalcik., "Tools and techniques for video performance

evaluation." In Proceedings 15th International Conference on Pattern Recognition. ICPR2000, 2000. 17. Kasturi, R., et al., "Framework for Performance Evaluation of Face, Text, and Vehicle

Detection and Tracking in Video: Data, Metrics, and Protocol." IEEE Transactions on Pattern Analysis and Machine Intelligence, 2009. 31(2): p. 319336. 18. Video Annotation Tool from Irvine, California. [cited 2017 28 July]; Available from: http://carlvondrick.com/vatic/.

156

Operational Evaluation of Do Not Block the Box Campaigns in Georgia
19. 24.1. Tkinter--Python interface to Tcl/Tk. [cited 01 August 2017]; Available from: https://docs.python.org/2/library/tkinter.html.
20. python-vlc 1.1.2. [cited 2017 1 August]; Available from: https://pypi.python.org/pypi/python-vlc/1.1.2.
21. VLC media player. [cited 2017 1 August]; Available from: https://www.videolan.org/vlc/. 22. Saroj, A., et al., "Analysis of Vehicle Blocking Behavior on Intersection Performance." In
TRB 96th Annual Meeting Compendium of Papers. 2017. 15 p. 23. Harris, D.S., "Traffic Management Alternatives for Business Improvement Districts." In
School of Civil and Environmental Engineering, 2014, Georgia Institute of Technology. p. 220. 24. Barry, C.D., "Calibration of The HCM 2010 Roundabout Capacity Equations for Georgia Conditions." In School of Civil and Environmental Engineering, Georgia Institute of Technology: Georgia Institute of Technology. p. 143. 25. Peesapati, L.N., "Methods To Evaluate The Effectiveness of Certain Surrogate Measures to Assess Safety of Opposing Left-Turn Interactions." In Civil and Environmental Engineering, 2014, Georgia Institute of Technology. p. 340.
157