|
Title |
AI and Computer Vision-Based Road Marking Condition Assessment
|
|
Authors |
강기상(Gi-Sang Kang) ; 유용래(Yong-Rae Yu) ; 안호준(Hojune Ann) ; 이종재(Jong-Jae Lee) |
|
DOI |
https://doi.org/10.11112/jksmi.2026.30.2.17 |
|
Keywords |
노면표시 상태 평가; 물리적 결함 분석; 객체 탐지; YOLOv8; IPM; 도로 유지관리 Pavement marking condition assessment; Deterioration analysis; Object detection; YOLOv8; Inverse perspective mapping (IPM); Road maintenance |
|
Abstract |
Conventional pavement marking inspections rely heavily on manual visual assessment, which is inherently subjective, labor-intensive, and inefficient for large-scale road networks. To overcome these limitations, this study proposes an AI-based system for quantitative analysis of deteriorations in road markings by integrating object detection and image processing techniques. The proposed system consists of five stages: data acquisition, preprocessing, object detection, deterioration analysis, and result reporting. Forward-facing road images captured by a vehicle-mounted camera are first transformed into a top-view representation using inverse perspective mapping (IPM) to eliminate geometric distortion and ensure spatial consistency. A YOLOv8-based object detection model is then employed to robustly extract lane marking regions under diverse road conditions. For quantitative defect assessment, boundary refinement using Harris corner detection and pixel-level density analysis is applied to estimate deterioration ratios, including micro-cracks and material peeling within the lane markings.Experimental validation using a small-scale testbed and real-world driving data demonstrated that the proposed system achieved a mean Average Precision (mAP) of 90.3% for lane detection and an average defect analysis accuracy of 86.4% compared to ground-truth measurements. These results confirm the effectiveness and reliability of the proposed system for automated road marking condition assessment.
|