Precision Road Infrastructure Management: Monocular Vision-Based 3D Damage Detection and Assessment

The study develops YOLO9tr, a lightweight object detection model for real-time pavement damage detection. The system identifies seven types of damage: lateral wheel marks, lateral cracks, alligator cracks, patching, potholes, crosswalk blur, and white line blur.

A grayscale-based depth estimation technique is also developed to estimate damage dimensions at distances of 3.5–7 meters without requiring specialized depth sensors. Experimental validation achieved a Mean Absolute Percentage Error (MAPE) of 19.4% for damage size estimation, while the proposed model can process images at speeds of up to 136 FPS.

The approach can operate with standard automotive-grade cameras and provides detection and measurement capabilities that can support road condition assessment, maintenance planning, resource allocation, and road infrastructure management.