Automated extraction of urban trees from mobile LiDAR point clouds

Individual Extraction of Street Trees From MLS Point Clouds Based on Tree Nonphotosynthetic Components Clustering

“…To verify the advantages of our method, we compared our extraction accuracy with that of other existing methods, including Wu2013 [13], Wu2016 [14], Li2016 [15], Teo2016 [11], Zhong2017 [12], Xu2018 [6], Ning2022 [9], and Li2022 [17]. Since the individual extraction results of street trees are highly related to the complexity of the scene where they are located, for objective comparison, we map the scene type of each study to the four common scenes defined in this article according to the street tree scene complexity presented in it.…”

“…Therefore, both of them are not competent for common Scene 4 where there is dense vegetation under the trees and the tree sizes vary greatly. Among other methods for comparison, Wu2013 [13], Wu 2016 [14], and Li2016 [15] all use the trunk detection-based method to segment and extract individual trees, while Li2022 [17] identifies and locates each tree based on the junction of trunk and branches, and Xu2018 [6] and Ning2022 [9] both use treetops as the basis for the individual segmentation of each tree after classification. Here, to compare the extraction results on complex common Scene 4, we select Wu2013 [13], Li2022 [17], and Xu2018 [6] as representative methods for street tree extraction based on trunk, trunk-branches junction, and treetop detection to individually segment and extract street trees in complex Road 6.…”

“…In recent years, many street tree individual extraction methods orient to MLS point clouds have been proposed. According to the different extraction strategies, these methods can be divided into classification based methods [6], [7], [8], [9], top-down segmentation methods [10], [11], [12], and bottom-up segmentation methods [13], [14], [15], [16], [17]. In the classification based methods, the raw point cloud is first classified by machine learning or deep learning, and then the tree points obtained by classification are individually segmented via certain segmentation or clustering algorithms (e.g., graph based segmentation [7] and mean shift based clustering [8]).…”

“…Compared with the method of classifying trees first and then segmenting them individually, this method of separating trees individually from objects containing nontrees has higher requirements of the segmentation algorithms. The bottom-up strategy first identifies and locates the tree positions (represented by tree trunk [13], [14], [15], tree trunk and branches [16], or the junction of trunk and branches [17]) through local features from the remaining objects after preprocessing, and then individually segments the tree points via location-based region growing [13], [14], [15] or clustering [16], [17].…”

Change Detection of Urban Trees in MLS Point Clouds Using Occupancy Grids

“…In Zhong et al (2013), the growing uses seeds that were selected by the local maxima in the horizontal histogram of the voxelised point cloud by octree decomposition. Using the improved supervoxel structure, in Wu et al (2016), a breadth-first search-based growing is employed, in which the seeds are selected by the gravity center of the supervoxel, which contains concentrated points. However, due to the scanning pattern of MLS, the identification of optimised seeds for growth is always a challenging task since either tree trunks or crowns suffer from occlusions, with merely partially scanned points acquired.…”