Road Damage Detection and Physical Properties Estimation

Overview

This project is oriented to the needs of urban municipal road traffic infrastructure maintenance, to carry out research on urban municipal road asset verification technology and urban municipal road pavement damage detection technology, and to select a typical urban municipal road section in Qiaokou District, Wuhan City, to carry out demonstration applications.

My Role

I am the student leader, responsible for the development of road damage detection algorithms, project schedule control, and also responsible for the coordination of all parties involved.

Research Content

1. A road damage dataset was constructed, which contains the physical properties (width, length, depth) of road damage.
2. The YOLOv5 architecture was enhanced by integrating additional regression heads to predict the physical properties of detected road damage.
3. An integral transform function (CDT) was developed to address the long-tailed distribution problem in values of regression variables.
4. The model was deployed on sanitation vehicles in Qiaokou District, Wuhan, China, to validate the algorithm’s robustness in real-world scenarios.