Abstract
Abstract. This paper proposes a method for constructing 3D models of shrubs with high accuracy from 3D point clouds acquired by terrestrial laser scanning (TLS). Since the shrub point cloud obtained by LiDAR scanning contains a large amount of redundant data, the focus of this method is to segment the branches and leaves of the shrub point cloud first, which can remove the noise and make the branch skeleton of the shrub stand out. Secondly, a triangulation network is constructed for the segmented branch points, and a minimum spanning tree (MST) is established from the triangulation network as the initial shrub skeleton of the input point cloud. Then, the redundant branches are removed by merging adjacent points and edges to simplify the initial skeleton. Finally, the 3D model of a shrub is constructed by a cylindrical fitting algorithm based on robust principal component analysis (RPCA). Experiments on different types of shrubs from different data sources demonstrate the effectiveness and robustness of the proposed method. The 3D models of shrubs can be further applied to the accurate estimation of shrub attributes and urban landscape visualization.