融合超体素与成对链接聚类的LiDAR点云分割算法  

LiDAR point cloud segmentation algorithm based on supervoxel and pairwise linkage clustering

作　　者：普东东 丁海勇 PU Dongdong;DING Haiyong(Nanjing University of Information Science&Technology,School of Remote Sensing&Geomatics Engineering,Nanjing 210044,China)

机构地区：[1]南京信息工程大学遥感与测绘工程学院,江苏南京210044

出　　处：《测绘通报》2022年第3期65-69,共5页Bulletin of Surveying and Mapping

基　　金：国家自然科学基金(41571350)。

摘　　要：针对现有的LiDAR点云分割算法稳健性差、效率低的问题,本文提出了一种新的层次化聚类分割算法。该算法首先把点云生成自适应分辨率的超体素,然后以超体素为基元,改进成对链接的分割算法,实现三维点云的分割。试验结果表明,该分割算法与现有的分割方法相比,具有更好的稳健性和更高的计算效率,避免了点云过分割和欠分割的问题。本文算法在分割细节方面更加突出,分割结果可有效地保证后续数据处理工作的精度。Aiming at the problems of poor robustness and low efficiency of existing LiDAR point cloud segmentation algorithms,this paper proposes a new hierarchical clustering segmentation algorithm.Firstly,a supervoxel with adaptive resolution is generated from the LiDAR point clouds.Then an improved pairwise linkage segmentation algorithm is used to the supervoxel to get the segmentation results.Experimental results show that the proposed segmentation algorithm has better robustness and higher computational efficiency compared with that of the existing segmentation methods.The issues of over segmentation and insufficient segmentation of the point clouds have been solved.The proposed algorithm is more prominent in segmentation details,and the segmentation results can effectively ensure the accuracy of subsequent data processing.

关 键 词：LIDAR点云 超体素 成对链接 分割 稳健性 

分 类 号：P23[天文地球—摄影测量与遥感]