Lane Marking Detection Using Mobile Mapping System Technology with Clustering Technique

Abstract

The sensitivity of cameras to lighting and environmental conditions affects the accuracy of lane detection in Autonomous Driving System. This study develops a lane detection technique using LiDAR data from a Mobile Mapping System, leveraging the intensity differences of lane markings painted with retro-reflective materials to enhance detection accuracy. The study covers a 2.8 km area around Shalun High-Speed Rail Station, Taiwan, with data processing conducted using MATLAB. Point cloud data is segmented to extract ground surface using the RANSAC algorithm, followed by lane detection using Spectral Clustering, which classifies data groups based on graph theory with Precision 90.70%, Recall 83.37%, F1-Score 86.88% and IoU 76.80%. The research show that Spectral Clustering enhances Autonomous Driving System accuracy overcoming camera limitations and improving transportation.