Point cloud features suitable for automatic labeling of MMS point cloud data

Abstract

For road mapping, it is important to add labels to point clouds captured by the Mobile Mapping System (MMS). Some automatic labeling methods have been proposed so far. However, in our experiment, conventional labeling methods were not sufficiently accurate for actual point clouds measured in Japan. In this paper, we propose a high-performance classification method that combines the multi-scale features of point clouds, the MMS specific features and the features obtained from point clouds mapped on the 2D image. The accuracy of the proposed method was evaluated using actual MMS data, and it was confirmed that the proposed method could achieve high recognition rate generalization performance. Our method can improve the accuracy of automatic labeling of point clouds, and is expected to improve the efficiency of map maintenance, which is a social infrastructure.