基于改进欧式聚类的工件分割方法  

Workpiece Segmentation Method Based on Improved Euclidean Clustering

作　　者：姜立豪 刘星桥[1] 李长峰 陈辉 马腾 赵德安[1] JIANG Lihao;LIU Xingqiao;LI Changfeng;CHEN Hui;MA Teng;ZHAO Dean(School of Electrical and Information Engineering,Jiangsu University,Zhenjiang 212013,China;Changzhou Mingseal Robot Technology Co.,Ltd.,Changzhou 213164,China)

机构地区：[1]江苏大学电气信息工程学院,江苏镇江212013 [2]常州铭赛机器人科技股份有限公司,江苏常州213164

出　　处：《软件导刊》2023年第11期161-167,共7页Software Guide

基　　金：国家自然科学基金项目(31571571,61903288)。

摘　　要：针对多工件检测中的单工件分割环节出现的欠分割和过分割问题,提出一种基于相邻工件区域识别的改进欧式聚类分割方法。首先将采集的点云直通滤波以获得感兴趣区域;然后基于离群点判定、法向量夹角和密度聚类的区域识别算法识别相邻工件区域;最后结合基于K近邻的自适应欧式聚类分割点云。结果表明,所提方法的分割精度在95%以上,能准确从场景中分割出单工件的感兴趣区域,并且分割效果较为稳定,可为后续的工件缺陷检测工作奠定基础。A modified Euclidean clustering segmentation method based on adjacent workpiece region recognition is proposed to address the issues of undersegmentation and oversegmentation in single workpiece segmentation in multi workpiece detection.Firstly,filter the collected point cloud directly to obtain the region of interest;Then,a region recognition algorithm based on outlier detection,normal vector angle,and density clustering is used to identify adjacent workpiece regions;Finally,combined with K-nearest neighbor based adaptive Euclidean clustering for point cloud segmentation.The results show that the proposed method has a segmentation accuracy of over 95%,can accurately segment the region of interest of a single workpiece from the scene,and the segmentation effect is relatively stable,which can lay the foundation for subsequent workpiece defect detection work.

关 键 词：机器视觉 欧式聚类 点云分割 聚类分割 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]