Traffic Sign Classification Method Using Corrected Intensity and Up-Sampled Point Cloud

Light detection and ranging (LiDAR), a crucial supplementary perception sensor for traffic sign classification (TSC) in intelligent transport systems (ITSs), offers 3-D point cloud impervious to ambient light and environmental changes. To address the potential camera failures in TSC, a novel LiDAR data-driven classification method was proposed to identify traffic sign patterns based on intensity correction (IC) and point cloud up-sampling (UP). The corrected intensity and up-sampled point cloud were employed to provide the reliable and accurate data sources for smart classification model training. An extensive series of experiments was carried out involving typical traffic sign patterns with diverse shapes and color designs. Results revealed a remarkable increase in the ${F}1$ score, a typical classification evaluation metric, from 0.6264 to 0.8153 after UP, and from 0.8153 to 0.8403 after IC. With similar density levels, a comparison of the classification results between the point cloud and images raised the requirement of developing multi-wavelength LiDAR, with high spatial resolution, high stability, and low divergence angle of the laser. The proposed method presents a viable and alternative solution for TSC in autonomous vehicles (AVs), paving the way for further advancements in ITS.