Applying AI to data sources to improve driver-pedestrian interactions at intersections

Tragically, roadway crossings sometimes act as killing fields, especially for pedestrian-vehicle collisions, with intersections accounting for 40% of transportation crashes in the US. This emphasizes the importance of effectively addressing pedestrian safety at intersections to mitigate such incidents. The first chapter of this report incorporates pedestrian safety into the optimization of traffic signals by collecting and linking data from traffic signals (cameras) and  analyzing the behaviors of pedestrians and drivers at intersections using Artificial Intelligence techniques, i.e., a decentralized Dyna Q-Learning environment. The results indicate that AI agents may safely prioritize pedestrian service even with longer waiting times or reduce pedestrian delays at the expense of vehicle delay performance. The report’s second chapter explores rare pedestrian crashes at intersections, called “corner cases,” using Fatality Analysis Reporting System (FARS) data and applying text analytics and the K-means unsupervised learning approach. Such crashes are likely to be triggered by a combination of factors, including poor visibility, severe weather, impaired pedestrian or driver behaviors, and dark lighting conditions. The final chapter of the research investigates the determinants of nighttime pedestrian crash injury severity in pedestrian-involved crashes on intersections using the Random Forest algorithm and ordered logit models. The analysis results reveal that alcohol impairment, foggy weather, elderly pedestrians, a speed limit of 50-55 mph, and motorists not yielding to pedestrians are more likely to contribute to severe pedestrian injuries at intersections. The implications of the findings are discussed in each chapter.