Integrating AI Video Analytics and Simulation for Surrogate Safety Analysis of Autonomous Vehicles at Urban Intersections

Authors:
Open Access
Publisher website
Google Scholar
Abstract
Advances in artificial intelligence and computer vision have enabled the use of Surrogate Safety Measures (SSMs) for proactive traffic safety assessment. This study, conducted within the STREET21 project, investigates the safety impacts of autonomous vehicle (AV) integration at a signalized intersection in Thessaloniki, Greece. Four hours of video data were processed with GoodVision software, allowing precise detection of pedestrian and vehicle trajectories. The extracted data supported the development, calibration, and validation of a PTV VISSIM microsimulation model, used with the Surrogate Safety Assessment Model (SSAM) to evaluate conflict indicators, particularly Post-Encroachment Time (PET). Multiple scenarios examined AV penetration rates and driving behaviors (cautious, normal, aggressive). Results indicate pedestrian–vehicle conflicts remain significant, especially during right-turn movements under amber phases. Increasing AV penetration improves safety outcomes, depending on driving behavior. The methodology demonstrates the robustness of combining AI-based video analytics with microsimulation for identifying traffic conflicts and assessing infrastructure readiness.

This publication has 18 references indexed in Scilit: