Invisible Road Hazard Detection Using YOLOv8 for Autonomous Vehicles

Accurate perception of road surface conditions is a critical requirement for autonomous driving systems. Hidden or partially visible road hazards such as potholes, waterlogging, faded lane markings, and surface debris significantly increase the risk of accidents and vehicle damage. This study presents a deep learning–based framework for multi-class road hazard detection using the YOLOv8 object detection model. A custom dataset comprising diverse road-hazard scenarios was constructed, annotated, and augmented to enhance model robustness across varying environmental conditions. The proposed approach was evaluated using standard performance metrics, including precision, recall, F1-score, and confusion matrices. Comparative experiments with YOLOv5s and YOLOv5L demonstrate that YOLOv8 consistently achieves higher detection accuracy and recall while maintaining realtime inference. The results confirm that YOLOv8 is a reliable and efficient approach for detecting invisible road hazards in autonomous vehicle applications.