Individual Extraction of Street Trees From MLS Point Clouds Based on Tree Non-Photosynthetic Components Clustering

The individual extraction of street trees from mobile laser point clouds is the prerequisite for their digital expression and application. However, due to the complexity of urban road environment and the diversity of street trees, especially the scenes where adjacent trees are different in type, size, and crown overlap, accurate extraction of individual street trees is still difficult to achieve. Therefore, in this paper, a new method to extract street trees individually from MLS point clouds is proposed. First, the ground and buildings are removed through data preprocessing. Then, the artificial poles that may overlap with street tree crowns are further removed by supervoxels region growing, and the regions of interest (ROI) including street trees and understory vegetation are selected. After that, the main branch part (including trunk) of each tree is separated from the ROI by non-photosynthetic components clustering. Finally, based on the individual clustering results of non-photosynthetic components, the remaining photosynthetic components in the ROI are segmented individually, and the vegetation under the tree is removed through gradual refinement to achieve the complete segmentation of individual trees. An urban area with a total road length of more than 2.1 km, including 6 roads with different complexity, was used to verify the effectiveness of the proposed method for the individual extraction of street trees. The results show that the proposed method can be effectively used for individual extraction of street trees in different complexity scenes. Overall, the precision, recall and F-score of street tree individual extraction are 94.5%, 97.4% and 95.9%, respectively.