641 Dilemma zone analysis using trajectory data acquired by video analytics

<h3>Background</h3> Numerous studies of driver behavior at signalized intersections and the associated safety impacts focused on the yellow light dilemma, whether to cross or not as the green light ends. New technologies provide exciting opportunities to enhance understanding of this issue, utilizing data at an unprecedented scale. In particular, when the operation of traffic signals is based on video analytics, trajectory data of all vehicles can be collected continuously and provide a vast basis for a thorough evaluation. <h3>Objective</h3> Examine the variation in driver behavior over time and between locations in terms of the yellow light dilemma. <h3>Methods</h3> The research discussed here relies on approximately ten weeks of data containing 21 million anonymous vehicle trajectories obtained by video analytics at 29 approaches in 10 intersections. We reduced the original trajectory data to critical features, including the vehicles’ speed and distance from the stop line at the onset of the yellow light and the drivers’ cross/stop decision. We examined alternative multivariate logistic regression models for the cross/stop decision in the straight direction at each approach and found that the best-fit model (in terms of BIC) combines the vehicles’ speed and distance from the stop line at the onset of the yellow light, and nighttime indicator (19:00–07:00). <h3>Results</h3> We found significant differences in behavior between approaches and intersections. It is particularly evident in the dilemma zone’s variability between approaches. For vehicles approaching an intersection at 15 m/s during daytime, the length of the dilemma zone ranges from 15 to 50 meters. <h3>Conclusions</h3> Variation between locations should be considered in signal timing procedures. Advanced video analytics technologies can be useful in understanding driver behavior and improving road safety.