Vehicle-Mounted Road Damage Detection Method Using YOLOv5s Framework
Wenjie Qian, Si Chen, Youxiang Huang, Xiaohu Xu, Feng Shi, Hong Wan, Zhiyu Lu · 2023
Timely detection and accurate recognition of road damage are crucial for road maintenance and traffic safety. Traditional road damage inspection methods, such as manual inspection and manual annotation, are inefficient and costly. Therefore, computer vision methods have become a faster and more cost-effective choice. In this study, we collected a dataset called RoadBHD, which consists of 4,579 images and 5,005 labels, including six types of road damage. We built a low-cost and efficient vehicle-mounted road damage detection system based on the YOLOv5s network model. Through experimental comparisons, the YOLOv5s network demonstrated superior performance in road damage detection compared to Faster R-CNN and SSD networks. This validates the feasibility of our designed vehicle-mounted road damage detection system.