Real-Time Vision-Based Detection and Severity-Level Classification of Road Damage

Road damage detection systems are critical for enhancing the safety and comfort of autonomous driving systems. Despite their importance, there has been limited research specifically focused on road damage detection in the context of autonomous driving. This study seeks to address this gap by developing and evaluating a computer vision-based road damage detection model utilizing the YOLOv11 architecture. The model was designed to classify road damage into three severity levels: low, medium, and high. Training was conducted using a self-collected dataset comprising 4,000 images captured from both urban and rural roads across Indonesia. Performance evaluation was carried out through simulations within the CARLA Simulator environment and real-time testing using recorded road videos. The results demonstrated that the model has an F1-score of 0.520 and could detect and classify road damage in both simulation and real-world settings in real time. These findings highlight the potential of deep learning techniques to enhance the perception systems of autonomous vehicles, particularly in complex and variable road conditions such as those found in Indonesia.