点云数据精简与配准研究  

Research on Simplification and Registration of Point Cloud Data

作　　者：庞正雅 周志峰[1] 钱莉[1] 叶珏磊 PANG Zheng-ya;ZHOU Zhi-feng;QIAN Li;YE Jue-lei(College of Mechanical Engineering,Shanghai University of Engineering Science,Shanghai 201620,China)

机构地区：[1]上海工程技术大学机械工程学院

出　　处：《软件导刊》2019年第6期25-28,34,共5页Software Guide

基　　金：上海工程技术大学研究生科研创新项目(18KY0105)

摘　　要：由于激光雷达等扫描设备得到的点云存在数据量大、数据中掺杂噪声较多等一系列问题,提出一种基于特征点保持的点云精简与配准方法。首先利用K-means算法对所有点云数据聚类,滤除掉噪声点云,再进行精简化处理;随后在精简的基础上用KD-tree对数据进行最近邻搜索以加快对应点查找速度,从而为配准节省一定的时间;最后根据欧氏距离选择合适的初值减少匹配误差。实验结果表明,精简后的点云数据保持了基本特征,一定程度上减少了配准时间和误差。Because of the point cloud obtained by scanning equipment such as laser radar has a series of problems,such as large data volume and high doping noise in data,a point cloud simplification and registration method based on feature point preservation is pro-posed. Firstly,K-means algorithm is used to cluster all point cloud data,filter out noise point cloud,and then simplify the processing. Then,on the basis of simplification,KD-tree is used to search the nearest neighbor of data to speed up the search speed of correspond-ing points,so as to save some time for registration. Finally,the matching error is reduced by selecting the appropriate initial value ac-cording to Euclidean distance.The experimental results show that the simplified point cloud data maintain basic characteristics and re-duce the registration time and error to some extent.

关 键 词：K-MEANS聚类 KD-TREE 点云精简 点云配准 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]