Automated traffic signs condition monitoring using machine vision

Abstract This study presents a unified multi-sensor framework for automated traffic sign condition monitoring, with a specific focus on tilt and occlusion detection under real-world roadway conditions. The proposed method integrates state-of-the-art image segmentation and object tracking to persistently identify signs across video frames, followed by two dedicated condition-assessment modules. The first module introduces an oriented bounding box (OBB)-based image-plane tilt estimation approach, which is physically validated using co-registered mobile LiDAR measurements to establish geometric consistency. The second module quantifies occlusion through mask coverage analysis to detect partial sign obstruction. Comprehensive field evaluation across diverse roadway environments demonstrates strong performance, achieving 92% accuracy in detecting tilted signs and 85% accuracy in classifying misaligned and occluded signs while maintaining real-time processing capability. The LiDAR-validated OBB-based tilt estimation and multi-sensor verification framework strengthen the methodological rigor of the approach. The proposed system provides a scalable and operationally deployable solution for continuous traffic sign inventory and condition assessment, supporting data-driven asset management for transportation agencies.