Abstract Manager

Routine traffic sign inspections that are conducted as part of highway maintenance and asset management programs are labor-intensive inspections prone to human error. In an attempt to automate the process, this study introduces a novel machine learning algorithm that works on detecting changes in traffic sign assets using low-density LiDAR data. Targeting efficient and cost-effective roadside asset management, the method simplifies the identification of damaged or missing traffic signs by comparing baseline and maintenance scans. The approach starts by restructuring the point cloud into a 3D grid through voxelization which helps enhance the detection accuracy. The next step involves implementing radial search with KD-tree structuring for efficient change detection, and employing the DBSCAN algorithm for false negative elimination. The method was tested across three different highways covering 15 km on which realistic damages within LiDAR point clouds was simulated including partial breakage and rotational changes. Testing demonstrated high accuracy, with F1 scores ranging from 92% to 100%. The method also proved exceptionally efficient, averaging just 115 seconds per kilometer in unattended processing. The method's adaptability enhances the management and maintenance of various highway elements. By addressing gaps in current traffic sign monitoring practices, this method offers a comprehensive, automated, and cost-effective solution for maintaining traffic signs and other highway infrastructure. This marks a significant advancement in traffic sign asset management and contributes substantially to the field of highway maintenance and safety.