Deep Learning Based Image Detection Model for Pavement Cracks and Potholes

Damaged road surfaces can cause traffic accidents and negatively affect the tires and suspension of vehicles, potentially contributing to accidents. Rapid detection is needed to prevent this. To overcome the limitations of human inspection of hundreds of thousands of kilometers of roads, we use CNN-based deep learning models. In this study, we used public big data from Kaggle site to detect road defects such as potholes and road cracks. We trained two deep learning models, YOLO v8 and a custom model, on approximately 58,000 images. For potholes, we identified their location and presence in the images and achieved a performance index mAP50 of 0.754. For road cracks, we were able to detect their presence or absence in the images with an accuracy of 84.96%. The proposed model can detect not only potholes and road cracks, but also road flooding, excessive curvature, and wear and tear.