A Computer Vision Approach to Evaluating Crosswalk Safety for Vulnerable Road Users

Abstract With advancements in computer vision and cloud computing, Surrogate Safety Measures (SSMs) now provide actionable insights to mitigate safety concerns before collisions occur. This study, conducted as part of the STREET21 research project and contributes to the existing body of knowledge by examining Post-Encroachment Time (PET), a time-based SSM, at a high-traffic urban intersection which many young vulnerable users (university students) cross as pedestrians for their daily commuting needs. In total 513 traffic conflict events were identified and mapped for the purposes of the analysis. The spatial analysis provides critical insights into the patterns of traffic conflicts. Results of the quantitative analysis demonstrate that pedestrian conflicts predominantly involved right-turning vehicles, followed by through vehicles, potentially indicative of red-light violations. The applied methodology underscores the efficacy of video analytics as a scalable alternative to traditional crash data analysis, enabling the evaluation of intersection designs and temporary treatments before permanent implementation.