Comprehensive Automated 3D Urban Environment Modelling Using Terrestrial Laser Scanning Point Cloud

In this paper we present a novel street scene modelling framework, which takes advantage of 3D point cloud captured by a high definition LiDAR laser scanner. We propose an automatic and robust approach to detect, segment and classify urban objects from point clouds hence reconstructing a comprehensive 3D urban environment model. Our system first automatically segments grounds point cloud. Then building facades will be detected by using binary range image processing. Remained point cloud will be grouped into voxels and subsequently transformed into super voxels. Local 3D features are extracted from super voxels and classified by trained boosted decision trees with semantic classes e.g. tree, pedestrian, and car. Given labeled point cloud the proposed algorithm reconstructs the realistic model in two phases. Firstly building facades will be rendered by ShadVis algorithm. In the second step we apply a novel and fast method for fitting the solid predefined template mesh models to non-building labeled point cloud. The proposed method is evaluated both quantitatively and qualitatively on a challenging TLS NAVTEQ True databases.