一种高效的点云去噪聚类方法  

An Efficient Method of Point Cloud Denoising Clustering

作　　者：钟文彬[1] 肖振远 刘光帅[2] ZHONG Wen-bin;XIAO Zhen-yuan;LIU Guang-shuai(The 10^(th)Research Institute of China Electronics Technology Group Corporation,Sichuan Chengdu 610036,China;School of Mechanical Engineering,Southwest Jiaotong University,Sichuan Chengdu 610031,China)

机构地区：[1]中国电子科技集团公司第十研究所,四川成都610036 [2]西南交通大学机械工程学院,四川成都610031

出　　处：《机械设计与制造》2022年第8期233-237,共5页Machinery Design & Manufacture

基　　金：国家自然科学基金资助项目(51275431);中国电子科技集团公司第十研究所技术创新基金(十所计20181218)。

摘　　要：逆向工程采用激光扫描设备获取点云数据时会引入无关离体噪声点,针对密度聚类法去噪时引起的内存占用较大、收敛时间较慢的问题,提出了一种高效的点云去噪聚类方法。首先,采用KD-tree建立点云数据的空间索引结构,选择点云的某一维度,找出最小索引核心点云。然后,沿着该维度方向在核心点云邻域外选择最近的未标记的点云扩展密度类。最后,分离出密度相连最大的点云类以完成点云去噪。实验结果表明,与统计滤波、半径滤波、密度聚类滤波算法相比,该方法在完整保留目标点云的同时能高效地去除离体点云噪声,提高了点云去噪的质量。Reverse engineering uses laser scanning equipment to acquire point cloud data,which introduces extraneous isolated noise points.Aiming at the problems of large memory consumption and slow convergence time caused by density clustering denoising,an efficient point cloud is proposed.Noise clustering method.First,the KD-tree is used to establish the spatial index structure of the point cloud data,and a certain dimension of the point cloud is selected to find the smallest index core point cloud.Then,along the dimension direction,the nearest unlabeled point cloud expansion density class is selected outside the core point cloud neighborhood.Finally,the point cloud classes with the highest density are separated to complete the point cloud denoising.Experimental results show that,compared with statistical filtering,radius filtering,and density clustering filtering algorithms,this method can effectively remove the noise of the in vitro point cloud while completely retaining the target point cloud,and improves the quality of point cloud denoising.

关 键 词：噪声点 密度聚类 KD-TREE 核心点云 密度类 密度相连 

分 类 号：TH16[机械工程—机械制造及自动化] TP391.41[自动化与计算机技术—计算机应用技术]