Using Direct-Georeferencing with a MEMS-LiDAR to Classify Traffic Signs, an Automatic Approach

In this article, a micro-electromechanical systems light-detection and ranging (MEMS-LiDAR) based sampling method for traffic sign recognition is presented. MEMS-LiDAR is an emerging remote-sensing technology in the autopilot (AP) area. Existing published scientific reports have been less concerned with using point-cloud datasets sampled from MEMS-LiDAR, as the MEMS-LiDAR is a recently matured commercial technology. In this paper, we propose utilizing a MEMS-LiDAR system (Robosense-M1) for data acquisition in a mobile laser scanning (MLS) in an urban environment. Subsequently, two methods related to clustering and point-cloud 3D geometric feature extraction are introduced to segment and filter out plain-like objects. Finally, a support vector machine method (SVM) is investigated for the task of traffic-sign classification. Our results successfully demonstrate the effectiveness of the proposed method with K1-score based on the MEMS-LiDAR sampled data to be 83.33%. This result shows the effectiveness of the proposed method.