In recent years, autonomous driving techniques have steadily developed and attracted numerous interests from industry and academia. High definition maps (HD Maps) become extra aided information for autonomous vehicle. Owing to the efficiency of mobile mapping system (MMS), MMS is usually utilized for mapping tasks to collect accurate geospatial data nowadays. Afterward, the mapping process is used to be applied manually. However, it takes a lot of work and time. Therefore, the purpose of this study is to automatically extract lane boundary which is one of important elements for autonomous vehicle to generate HD Maps. Moreover, this study only focuses on processing point clouds containing 3D information of surroundings. In order to improve the result of lane boundary extraction, the study implements pre-processing include ground point clouds and non-ground point clouds segmentation and noisy point removal. This study conducts two algorithms to realize lane boundary extraction. One is using the voxelization algorithm, and the other is extract lane boundary based on Euclidean clustering algorithm. The correctness of lane boundary extraction can achieve 86.7% and 95.2% respectively. From the results, the extraction method based on Euclidean clustering algorithm is better and more flexible in lane marking extraction. The detailed analysis is given to assess the availability of proposed methods.

Keywords: point cloud, intensity, lane boundary extraction, voxelization, Euclidean clustering