Refined Identification of Vehicle Spatiotemporal Parameters Based on Multi-View Machine Vision Fusion

Accurate spatiotemporal parameter identification of vehicles is essential for precise vehicle load estimation. This study proposed a novel multi-view machine vision fusion method for refined vehicle spatiotemporal parameter identification. An improved CA-YOLOv5 deep learning model was constructed for vehicle and wheel detection. The coordinate attention mechanism module would improve the detection performance of the model under complex traffic scenes. The DeepSORT algorithm was introduced to track the trajectory of vehicles and wheels. Then, the trajectory was mapped from image coordinates to the practical bridge deck coordinate by direct linear transformation. In order to solve the problem of invisibility from a single view due to vehicle occlusion in a complex traffic scene, a multi-view fusion algorithm was developed through integrated analysis of video data collected by multiple cameras from different angles. A model experimental study was carried out to validate the robustness and accuracy of the proposed method. The results indicated that the introduction of the CA mechanism effectively improved the detection performance of the model under complex traffic scenes. The trajectory of each wheel could be accurately tracked by the proposed YOLO-DeepSORT model, and the multi-view fusion algorithm improved the identification accuracy and enhanced the tracking performance under complex traffic scenes such as vehicle occlusion.