Multi-object trajectory extraction based on YOLOv3-DeepSort for pedestrian-vehicle interaction behavior analysis at non-signalized intersections

Pedestrian-vehicle interaction behavior analysis at non-signalized intersections has become the important content of traffic safety research. In this paper, we focus on presenting a processing framework for the analysis of pedestrian-vehicle interaction behavior based on YOLOv3-DeepSort. Comparative experiments are done for verifying the performance of the presented framework. The presented YOLOv3-DeepSort can achieve the results that the ML value is 14.10% and the IDs value is 382 on the MOT16 dataset. And YOLOv3–416 can achieve the efficiency of 29 ms. Results show that the presented framework has excellent performance for trajectory extraction. Furthermore, the methodology is confirmed by the presented case. And we found that the studies on exit interactions can provide theoretical evidences for presenting some protective measures, to ensure the safety of pedestrians and vehicles at non-signalized intersections.