三维点云特征的工件识别与位姿估计  

Workpiece Recognition and Pose Estimation Based on 3D Point Cloud Features

作　　者：张铁[1] 肖卓 邹焱飚[1] 何英武 ZHANG Tie;XIAO Zhuo;ZOU Yan-biao;HE Ying-wu(School of Mechanical&Automotive Engineering,South China University of Technology,Guangdong Guangzhou 510640,China;Guangzhou CNC Equipment Co.,Ltd,Guangdong Guangzhou 510000,China)

机构地区：[1]华南理工大学机械与汽车工程学院,广东广州510640 [2]广州数控设备有限公司,广东广州510000

出　　处：《机械设计与制造》2022年第2期252-256,共5页Machinery Design & Manufacture

基　　金：国家04科技重大专项(2015ZX04005006);中山市科技重大专项资助项目(2016F2FC0006,2018A10018);广东省科技计划资助项目(2019B040402006)。

摘　　要：针对工业机器人在抓取工件过程中需要识别和定位工件的问题,提出了一种基于改进法矢计算的点云局部描述符SHOT的三维物体识别与位姿估计的方法。首先,对三维扫描仪获取工件表面点云进行预处理和分割,得到用于匹配的工件点云模板数据集;其次,采用均匀采样算法提取特征点集,通过SHOT特征描述符对场景点云与模板点云提取的特征点进行点云局部描述;最后,利用Kd树建立初始对应对,引入3D霍夫投票机制通过点云局部坐标系各向量之间的转换关系完成工件识别,根据对应关系求解出初始变换矩阵。基于ICP算法进行精确配准,得到工件的精确位姿信息。实验结果表明:该方法与基于ICP算法进行对比,提高了配准精度,可以得到精确的位姿估计信息,为在机器人应用中工件定位抓取提供精确的位姿检测方法。Aiming at the problem that robots need to identify and locate the workpiece in grasping process,a method for3 D object recognition and pose estimation based on SHOT,a local descriptor with improved normal vector of point cloud date,is proposed.First,the point cloud data of the workpiece surface obtained from 3 D scanner is preprocessed and segmented to obtain the point cloud dataset of workpieces for matching. Second,the uniform sampling algorithm is adopted to extract the feature point set,and SHOT descriptor is used to describe the feature points extracted from the scene point clouds and template point clouds locally. Finally,K-dimension tree is used to establish the initial correspondences,3 D hough voting mechanism is introduced to complete the workpiece identification through the transformation relationship between vectors in the local coordinate system,and the initial transformation matrix is solved according to the corresponding relationship. Based on Iterative Closest Point(ICP)algorithm,accurate registration is carried out and accurate pose information of the workpiece is obtained. The experimental results show that compared with ICP algorithm,the proposed method improves the registration accuracy,and the accurate pose estimation information can be obtained,which provides an accurate pose detection method for positioning and grasping the workpiece in robot applications.

关 键 词：三维物体识别 3D霍夫变换 ICP算法 位姿估计 机器视觉 

分 类 号：TH16[机械工程—机械制造及自动化] O436[机械工程—光学工程]