一种面向分拣的改进欧式聚类分割算法  

An Improved European Clustering Segmentation Algorithm for Sorting

作　　者：杨诚轩 梁庆华[1] 罗磊[1,2] 孟凡雨 YANG Chengxuan;LIANG Qinhua;LUO Lei;MENG Fanyu(School of Mechanical Engineering,Shanghai Jiaotong University,Shanghai 200240,China;Shanghai Smart State Technology Co.,Ltd.Shanghai 201306,China)

机构地区：[1]上海交通大学机械与动力工程学院,上海200240 [2]上海交大智邦科技有限公司,上海201306

出　　处：《机械设计与研究》2022年第2期174-178,共5页Machine Design And Research

基　　金：上海市人工智能创新发展专项(2019-RGZN-01026)。

摘　　要：点云分割是机器人无序分拣的关键技术,主要难点是如何从工况中分割出完整的目标工件点云。常用方法中,RANSAC分割算法不能提取完整的螺栓点云,而传统的欧式聚类分割算法易受聚集噪点和点云缺失影响且耗时较长。为了解决以上问题,针对六角螺栓,提出一种基于欧式聚类和点法向量特征优化的点云分割算法。首先对三维点云数据进行体素化处理得到体素云,然后在体素空间内用RANSAC算法去除地面,利用八叉树(Octree)建立点云间拓补关系,然后根据K邻域搜索的欧式距离判别对点云进行快速的聚类分割,最后根据点法向量优化分割结果。实验结果表明本文的分割算法可以提高处理效率,使计算量大大减少,并且使最后的分割结果更加精确。Point cloud segmentation is a key technology for robots’disorderly sorting.The main difficulty is how to segment a complete target workpiece point cloud from the working space.Among the commonly used methods,the RANSAC segmentation algorithm cannot extract the complete bolt point cloud,while the traditional European clustering segmentation algorithm is susceptible to the effect of clustered noise and missing point clouds and takes a long time.In order to solve the above problems,this paper proposes a point cloud segmentation algorithm based on European clustering and point normal vector feature optimization for hexagonal bolts.First,the 3 D point cloud data is voxelized to obtain the voxel cloud;then the RANSAC algorithm is used to remove the ground in the voxel space;Octree is used to establish the topological relation between the point clouds,and then the K neighborhood search with Euclidean distance discrimination is performed for fast clustering and segmentation of the point cloud.Finally,the segmentation result is optimized according to the point normal vector.Experimental results show that the segmentation algorithm in this paper can improve processing efficiency,significantly reduce the amount of calculation,and make the final segmentation result more accurate.

关 键 词：点云分割 机器视觉 欧式聚类 RANSAC 

分 类 号：TP242.2[自动化与计算机技术—检测技术与自动化装置]