Traffic conflict analysis in continuous confluence area of cross-river bridge driven by vehicle trajectory data.

Objectives: To conduct an in-depth study on the spatial distribution of traffic conflicts in the continuous merging areas of cross-river bridges and ensure public transportation safety.

Methods: First, we utilized drone aerial photography to collect videos of vehicle movements. Using the YOLOv7 object detection algorithm and the Strong SORT multi-object tracking algorithm, we extracted high-precision vehicle trajectory time-series data. Next, based on the motion characteristics of traffic entities, we proposed using Deceleration Rate (DR) to describe rear-end conflicts and Lane Change Speed (LCS) to describe lane-changing conflicts. Additionally, we employed the K-means clustering method to determine the threshold values for minor, moderate, and severe levels of rear-end and lane-changing conflicts. Finally, based on the obtained trajectory data, the values of traffic conflicts are calculated and their severity is classified. A heat map of the spatial distribution of vehicle conflicts in continuous merging zones is then created to study the spatial distribution patterns of traffic conflicts.

Results: The threshold values for minor, moderate, and severe levels of rear-end conflicts are determined to be 3.06 m/s, 5.36 m/s, and 8.04 m/s, respectively. For lane-changing conflicts, the thresholds are 1.13 m/s, 2.07 m/s, and 3.45 m/s. The spatial distribution of traffic conflicts exhibits a "first increase, then decrease, and then increase again" trend.

Conclusions: The study identifies the critical areas of traffic conflicts in the continuous merging zones of cross-river bridges. The research results provide a novel approach for acquiring traffic data in these areas and offer a reliable quantitative method for assessing safety risks on these road segments. This provides a theoretical basis for proposing targeted traffic safety management strategies.