Edge-Ready Road Damage Detection Using An Enhanced YOLO With Hyperparameter Tuning

Abstract

Maintaining high-quality road infrastructure is essential for urban safety, efficient traffic flow, and reduced vehicle maintenance costs. Traditional manual inspections are time-intensive, costly, and prone to human error, highlighting the need for automated monitoring systems. Recent advances in deep learning and computer vision have made real-time road damage detection feasible. This study presents a lightweight road damage detection framework based on the YOLOv10n model, optimized for deployment on edge devices. The system integrates hyperparameter tuning to balance accuracy and computational efficiency. Experimental results show the framework achieves a precision of 0.986, recall of 0.973, mean average precision (mAP@0.5) of 0.988, and an F1-score of 0.978 for detecting cracks, potholes, and surface wear. Deployment on NVIDIA Jetson Nano achieved 7.5 FPS, and on NVIDIA AGX Orin, 67 FPS. These findings demonstrate that the proposed framework is suitable for scalable, real-time road monitoring, supporting proactive maintenance strategies and enhancing urban transportation safety.