Safety analysis of pedestrian–vehicle interaction behavior at rainy intersections via SFD-ResTrack

Abstract

The analysis of pedestrian–vehicle interaction behavior at rainy intersections, grounded in trajectory data, has become a pivotal aspect of contemporary traffic safety research. Detection-based tracking methods are widely recognized as effective means of extracting traffic trajectory data. Nevertheless, when applied to traffic surveillance videos, the performance of some methods remains constrained by variations in rain intensity and complex occlusion effects. In this paper, a spatial-frequency domain guided residual network with deep association tracking (SFD-ResTrack) framework is used to support the safety analysis of pedestrian–vehicle interaction behavior at rainy intersections. Experiments are conducted to demonstrate the effectiveness of SFD-ResTrack. For trajectory estimation in real-world rainy intersections, SFD-ResTrack achieves a root mean square deviation (‌RMSD) of 2.1 under diverse rainfall conditions. The case study verifies the applicability of the proposed methodology, and the results indicate that analyzing exit interactions can provide valuable insights for improving pedestrian and vehicle safety at rainy intersections.