点云特征提取精简方法研究  

Research on Simplification Method of Point Cloud Feature Extraction

作　　者：田赢 陈裕汉 吴学群[1] 刘洋[1] 韩啸 张豫宁 TIAN Ying;CHEN Yuhan;WU Xuequn;LIU Yang;HAN Xiao;ZHANG Yuning(Faculty of Land and Resources Engineering,Kunming University of Science and Technology,Kunming 650093,China;School of Earth Science and Engineering,West Yunnan University of Technology,Dali 671000,China)

机构地区：[1]昆明理工大学国土资源工程学院,云南昆明650093 [2]滇西应用技术大学地球科学与工程学院,云南大理671000

出　　处：《软件导刊》2023年第12期223-231,共9页Software Guide

基　　金：滇西应用技术大学人才引进科研启动项目(2021RCKY0005)。

摘　　要：为解决传统点云精简算法精简的点云特征易丢失、空洞较多等问题,提出基于邻域法向夹角和二分K-means聚类特征提取的点云精简算法。该算法利用点云邻域法向夹角提取出模型的整体特征点,将点云曲率作为二分Kmeans聚类算法的聚类特征,初步提取出模型的局部特征点;然后计算初步局部特征点的平均曲率,将大于平均曲率的点提取为最终的局部特征点,对模型进行特征保留,对非特征点则利用改进的体素精简法进行下采样;最后将提取出的整体特征点、局部特征点与非特征点合并,从而完成点云的简化。该算法与传统的精简法、其他文献中的方法相比,可获得更高的精度。In order to solve the problems of the traditional point cloud simplification algorithm,such as easy feature loss and more holes,a point cloud simplification algorithm based on neighborhood normal angle and dichotomous K-means clustering feature extraction is proposed.The algorithm uses the normal angle of the point cloud neighborhood to extract the whole feature points of the model.Using point cloud curva⁃ture as the clustering feature of dichotomous K-means clustering algorithm,the local feature points of the model were initially extracted.Then,the average curvature of the initial local feature points is calculated,and the points larger than the average curvature are extracted as the final local feature points,and the feature of the model is retained.For the non-feature points,the improved voxel reduction method is used for downsampling.The extracted global feature points,local feature points and non-feature points are combined to complete the simplification of point cloud.Compared with the traditional reduction method and other methods in literature,the proposed algorithm has higher precision.

关 键 词：点云精简 特征提取 邻域法向夹角 二分K-means聚类 点云曲率 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]