Abstract
Roadside tree segmentation and parameter extraction play an essential role in completing the virtual simulation of road scenes. Point cloud data of roadside trees collected by LiDAR provide important data support for achieving assisted autonomous driving. Due to the interference from trees and other ground objects in street scenes caused by mobile laser scanning, there may be a small number of missing points in the roadside tree point cloud, which makes it familiar for under-segmentation and over-segmentation phenomena to occur in the roadside tree segmentation process. In addition, existing methods have difficulties in meeting measurement requirements for segmentation accuracy in the individual tree segmentation process. In response to the above issues, this paper proposes a roadside tree segmentation algorithm, which first completes the scene pre-segmentation through unsupervised clustering. Then, the over-segmentation and under-segmentation situations that occur during the segmentation process are processed and optimized through projection topology checking and tree adaptive voxel bound analysis. Finally, the overall high-precision segmentation of roadside trees is completed, and relevant parameters such as tree height, diameter at breast height, and crown area are extracted. At the same time, the proposed method was tested using roadside tree scenes. The experimental results show that our methods can effectively recognize all trees in the scene, with an average individual tree segmentation accuracy of 99.07%, and parameter extraction accuracy greater than 90%.