Large field monitoring system of vehicle load on long-span bridge based on the fusion of multiple vision and WIM data

Abstract

Vehicle Load Monitoring (VLM) on entire long-span bridge decks presents significant challenges due to the spatial and temporal randomness of vehicles. Existing VLM systems often suffer from limited viewing coverage and poor continuity of multi-vehicle tracking methods. This paper proposes a VLM system consisting of multi-vision image pre-processing, modified YOLO-v4 model, kinematics-enhanced vehicle tracking algorithm, and data fusion method between vision and Weigh-In-Motion sub-systems. The system was tested on a long-span bridge using six cameras, achieving vehicle monitoring of entire deck. The precision of multi-vehicle tracking achieved 99.28% built upon YOLO-v4 model with 96.2% mean Average Precision (mAP). Comparative results demonstrate that the modified YOLO-v4 model outperforms state-of-the-art approaches, and our proposed tracking method surpasses other methods. Our proposed system offers a comprehensive solution for VLM on entire bridge deck, overcoming the limitations of existing methods. Future work could extend the system's capability to include complex traffic patterns.