Field Testing Demonstrated 93 Percent Accuracy in Pedestrian Detection and Dynamic Flash Yellow Arrow Signal Decisions With LiDAR Tracking.
Field Tests and Simulations of Dynamic Flash Yellow Arrows at Texas Intersections Evaluated Changes in Queue Length, Delay, and Pedestrian Conflicts.
Arlington, Texas, United States
Irving
Developing a Tracking-Based Dynamic Flash Yellow Arrow Strategy for Permissive Left-Turn Vehicles to Improve Pedestrian Safety at Intersections.
Summary Information
Flash yellow arrow (FYA) is a left-turn strategy at signalized intersections that allows left-turn vehicles to cross during gaps in opposing through traffic. However, it can raise pedestrian safety concerns, especially in certain conditions such as heavy traffic, blocked views, or darkness. An alternative, a FYA with a “minus-pedestrian” feature, suppresses the FYA for a cycle if the pedestrian phase is called but may also eliminate left-turn capacity beyond the needed pedestrian crossing time. This study proposed a dynamic flash yellow arrow (D-FYA) algorithm using a light detection and ranging (LiDAR) tracking system to protect crossing pedestrians from permissive left-turn vehicles and reactivate once cleared. It was tested in the field at an intersection next to the campus of the University of Texas at Arlington, as well as via simulation analysis.
METHODOLOGY
The D-FYA algorithm was developed to recognize pedestrians who entered the “wait zone” and pushed the call button. If the LiDAR detected pedestrians in the “boundary zone,” the algorithm suspended the programmed FYA temporarily, reactivated when the pedestrians left the “hazard zone.” The D-FYA algorithm was evaluated for reliability and accuracy in two case studies:
- Case Study I used an emulation-in-the-field framework at the intersection of Cooper Street and UTA Boulevard in Arlington, Texas. This intersection had a daily pedestrian crossing volume of 1,000 to 1,500. The real-time D-FYA decisions were verified over 100 traffic cycles, 70 of which had at least one pedestrian phase called. Whenever a phase started, the D-FYA algorithm was run and reported its decisions on a console screen, and it was then compared to field observations.
- Case Study II tested the D-FYA’s queue lengths, delay, and safety performance as opposed to two other permissive left-turn strategies: (1) protected + permissive left turn (PPLT) and (2) PPLT + minus-pedestrian-phase). The tests were conducted in nine traffic simulation modelling scenarios with different combinations of vehicle and pedestrian volumes (low, medium, and high). The simulation model was based on the intersection of West Walnut Hill Lane and North Belt Line Road in Irving, Texas.
The study noted the limitation of LiDAR in cases when pedestrians leaned on traffic light poles or multiple pedestrians stood too close for the algorithm to separate them effectively.
FINDINGS
- In Case Study I, comparing what the D-FYA reported on screen with field observations, the accuracy rate of the algorithm was 93 percent (93 of 100 cycles).
- In Case Study II, the queue length performance of D-FYA was between those of the two other left-turn strategies. A similar pattern was found in the delay analysis. However, when both vehicle and pedestrian volumes increased to high, all three strategies showed similar delays and queue lengths.
- In Case Study II, the number of conflicts caused by permissive left turns were reduced under the D-FYA compared to standard FYA. Low traffic volumes had 36 versus 58 conflicts. Medium traffic volumes had 676 versus 718 conflicts, and high traffic volumes had 730 versus 860 conflicts. For other safety metrics, the differences were not significant.
