A supervoxel approach to the segmentation of individual trees from LiDAR point clouds
Published in Remote Sensing Letters, 2018
Recommended citation: Xu Sheng, Ye Ning, Xu Shanshan(*) and Zhu Fa. "A supervoxel approach to the segmentation of individual trees from LiDAR point clouds." Remote Sensing Letters. vol. 9(6), pp.515-523, 2018, doi: 10.1080/2150704X.2018.1444286. https://www.tandfonline.com/doi/full/10.1080/2150704X.2018.1444286
This letter aims to propose an automatic method for the segmentation of street trees from LiDAR point clouds. The first step is to extract tree points to reduce the scale of data to be processed. The second step is to formulate supervoxels for grouping homogeneous points. The third step is to optimize the segmentation of individual trees based on the minimum distance rule. Each supervoxel will be assigned the same tree index as its spatially closest treetop supervoxel, therefore, points from the homogeneous region sharing the same tree index. Experiments show that we achieve the completeness of 78.5%, correctness of 94.5% and -score of 0.85 in the airborne LiDAR data. For mobile LiDAR data our accuracy is 100, 93.8% and 0.96 in terms of the recall, precision and Fscore, respectively.
Recommended citation: ‘S. Xu, N. Ye, S. Xu(*), F. Zhu. "A supervoxel approach to the segmentation of individual trees from LiDAR point clouds." Remote Sensing Letters. vol. 9(6), pp.515-523, 2018.’