RoadGuard: An Automated Road Damage Detection and Assessment System Using YOLOv8

Abstract

Road surface defects such as potholes and cracks can reduce driving safety and increase maintenance costs when they are not identified early. This paper presents RoadGuard, an automated road-damage detection and assessment prototype that combines YOLOv8-based object detection with interpretable severity and repair-priority analysis. The system accepts a road image through a Flask web interface, stores and previews the selected file, performs YOLOv8 inference, and processes each detected object using separate severity and priority modules. Severity is estimated from the relative bounding-box area and damage type, while repair priority combines severity, damage importance, and detection confidence. In a representative end-to-end test, the system detected one pothole with 84.2% confidence and two cracks with 38.4% and 27.2% confidence. Their computed severity scores were 26.03, 25.06, and 6.83, respectively, while the corresponding repair-priority scores were 53.25, 38.31, and 25.67. The prototype demonstrates how object-detection outputs can be transformed into maintenance-oriented information through a lightweight web application. The reported confidence values are per-detection inference scores, not overall model accuracy. The system is intended as a decision-support prototype and can be extended with segmentation, depth estimation, GPS mapping, larger benchmark evaluation, and field calibration. Index Terms—Road damage detection; YOLOv8; computer vision; pothole detection; crack detection; severity assessment; repair priority; Flask; intelligent transportation systems