Road Damage Detection and Classification Using YOLOv5
K C Rehana, G Remya
Abstract
Road infrastructure is a critical public asset that must be inspected and monitored regularly since it contributes to economic progress. However, road surface deteriorates over time from various factors. Detecting road damage quickly and precisely allows road-maintenance agencies to conduct timely repairs, maintain optimal road conditions, and improve transportation safety. For proper road maintenance, detection should be automated to eliminate time-consuming and inefficient manual inspection. The scope of automatic pavement defect assessments has significantly improved over the years. In this paper, we propose road damage detection and classification system that uses YOLOv5 to automatically detect damages from the road images. The model considers seven categories comprising mainly cracks , namely D00, D10, D20, D40, D43, D44 and D50. Evaluation results show that the overall accuracy of the road damage detection and classification model is 92%. We illustrate how YOLOv5, one of the most recent state-of-the-art object detection and classification algorithms, can be utilized to perform this task in a fast manner with effective results.