一种利用曲率约束的改进K-means三维点云数据分割方法  

Improved K-means Segmentation Method Using Curvature Constraints for Three-dimensional Point Cloud

作　　者：杨永涛 黄国言[2] 张坤[3] 吴培良[2] 

机构地区：[1]燕山大学信息技术中心,河北秦皇岛066004 [2]燕山大学信息科学与工程学院,河北秦皇岛066004 [3]河北科技大学信息科学与工程学院,河北石家庄050018

出　　处：《小型微型计算机系统》2017年第11期2573-2579,共7页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61305113)资助;河北省自然科学基金项目(2016203358)资助

摘　　要：为了提高三维点云逆向重建中对局部细节部位的敏感性,解决表面特征变化较大、外形较为复杂的点云数据分割不理想对后续处理产生较大影响的问题,提出一种利用曲率约束的三维点云数据分割新方法.该方法首先利用点云数据的坐标信息,计算出对应的曲率信息,然后基于坐标和曲率对点云之间的距离进行定义,在此基础上,按照K-means聚类的思想,实现点云的分割.同时,为了解决聚类分割对初始聚类中心的依赖,提高分割效率,提出一种基于立方体素栅格的点云初始聚类中心选取方法.实验结果表明,本文方法实现了点云数据特征明显部位的细分割,通过调整约束参数可以适用于曲面变化差异程度不同的点云数据分割,初始分割中心的选取方法保证了分割结果的唯一性和有效性,大大减少了消耗的时间,明显提高了效率,本文方法对实际应用具有积极的意义.Inthe pointclouds processing for reverse engineering,the segmentation plays an important impacton the subsequent processing,especially when the surface has more and complex characteristics. In order to get a feasible segmentation result and improve the sensitivity to the local detail parts for 3 D point cloud reconstruction,this paper presents a new method for 3 D point cloud segmentation,which is based on curvatureinformation. Firstly,calculate the curvature informationbased on the point coordinates,and thendefine the distance between pointsusing the coordinate and curvature information,at last,segment the point cloudaccording to the K-means clusteringtheory. Meanw hile,in order to avoid the dependency on initial centers and improve the efficiency,this paper also presents aselection method of initial clustering centers based oncube voxel grid. Experimental results show that the method in this paper can get a feasible segmentation result,by adjusting the parameter,it can applied to the point cloud with different surfaces,the selection method of initial centersensuringthe uniqueness and effectivenessofthe segmentation result,it can greatly reduce the time overheads,and achieve higher efficiency,the method in this paper is positive to practical applications.

关 键 词：点云分割 曲率 体素栅格 K-MEANS 初始中心 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]