Abstract
With the continuous advancement of autonomous driving technology, lane marking-based environment perception has become a critical component of autonomous vehicle systems. However, long-term vehicle loads cause road markings to deteriorate and fade, significantly compromising driving safety. Traditional road marking quality inspection methods are inefficient and struggle to achieve high-performance, convenient detection. To address these challenges, this paper proposes an integrated framework for road marking detection and evaluation using Unmanned Aerial Vehicle (UAV) imagery. The framework comprises three core modules: lightweight data acquisition, efficient marking extraction, and accurate distress assessment. First, optimized UAV flight parameters enable low-cost, highly flexible, and safe data collection. Second, the YOLOv8-MEB model, combined with instance segmentation screening and local image optimization, achieves lane segmentation precision and recall above 90% with FPS exceeding 60. Furthermore, a standard marking template library is constructed, and a RANSAC-based template matching method with affine transformation is employed to restore intact marking shapes. A contour correction strategy is introduced to mitigate errors induced by construction inaccuracies. The proposed framework supports nine common types of road markings and yields approximately 10% error in distress ratio calculation under non-severe damage conditions, providing a practical technical reference for intelligent road maintenance.