ABSTRACT
With the rapid advancement of vehicle load monitoring systems, incorporating dynamic weighing and multi-view visual technologies, acquiring precise vehicle loads across large fields of view on bridge surfaces has become crucial for bridge health monitoring. However, for ultra-long-span bridges, due to the limitations of single-camera focal length and depth of field, multi-camera and re-identification technologies are usually necessary to achieve full-bridge coverage. This approach inevitably leads to high system construction costs, large data transmission and processing costs, and compromises system robustness. To overcome these challenges, this paper proposes a vehicle load monitoring method for large-span bridges enhanced by super-resolution technology. The proposed approach introduces a clustering segmentation technique based on image clarity, enabling recognition and cropping of far-focal blur areas. Then, a super-resolution image enhancement network is designed to restore the resolution of blurry regions. Finally, a vehicle recognition algorithm based on YOLO is implemented to achieve wide-field vehicle recognition in single-view images. The research results demonstrate that super-resolution technology can effectively improve vehicle recognition accuracy and robustness within blurry areas of vehicle monitoring images, thus achieving wide-field bridge surface vehicle monitoring under single-view conditions. By significantly reducing the cost and complexity of vehicle monitoring systems, this method provides important value for bridge traffic management and real-time performance evaluation.
