RoadGuard: A real-time YOLO-based intelligent pothole detection and severity assessment framework for smart transportation infrastructur

Abstract

Road is the vital connection for communication between different places in our daily life. The periodical maintenance of road surface, particularly detection of potholes in a prior basis is very important and a major challenge not only for transportation safety but also important for preventing several accidents and human death. Conventional pothole detection methods depend heavily on manual or public surveys and reporting, which is a very delayed process of maintenance and inefficient monitoring. This paper proposes RoadGuard, an intelligent real-time pothole detection and severity assessment framework using YOLO-based deep learning models that help to create smart road infrastructure management by enabling the opportunities to establish secure communication through road. The proposed framework explores three different YOLO variants including YOLOv8n, YOLOv8s, and YOLOv8m for efficient real-time inference that integrates automatic pothole detection and counting, severity classification, and maintenance related record for smart city infrastructure management. Experimental observations indicate that the proposed approach improves automation, scalability, and practical deployment feasibility compared to traditional inspection methods. Results demonstrate that all three variants achieve mAP@0.5 above 89%, confirming reliable pothole detection under varied road and lighting conditions.