A Real-Time Pedestrian-Vehicle Collision Risk Warning Framework at Signalized Intersections Using Trajectory Prediction and Dynamic Conflict Thresholds

Abstract

Pedestrian-vehicle collisions remain a critical safety concern at urban intersections, where existing risk assessment methods often rely on static conflict thresholds that lack adaptability to dynamic traffic conditions. To address this gap, this study develops a real-time pedestrian–vehicle collision risk warning framework integrating Hidden Markov Model-based vehicle trajectory prediction, an improved minimum extended time-to-collision (ETTCmin) conflict indicator with dynamic thresholds derived via spline interpolation based on the real-time vehicle-to-pedestrian ratio, and a probabilistic conflict-based collision risk estimation model incorporating driver reaction and braking times. Validated using the SinD dataset and field data from Xi’an, the proposed method demonstrates robust trajectory prediction accuracy, earlier hazard detection compared to traditional time-to-collision metrics. By enabling adaptive, interpretable, and real-time risk warnings, this approach provides a proactive and scientifically grounded tool for enhancing pedestrian safety and supporting targeted traffic management interventions at intersections.