双目视觉空间定位系统中点云目标融合分割算法  

Point cloud target fusion segmentation algorithm in binocular visual spatial positioning system

作　　者：彭凯[1] 饶钰涵 肖艳军[1] 杨泽青[1] PENG Kai;RAO Yuhan;XIAO Yanjun;YANG Zeqing(School of Mechanical Engineering,Hebei University of Technology,Tianjin 300131,China;School of Precision Instrument and Opto-Electronics Engineering,Tianjin University,Tianjin 300072,China)

机构地区：[1]河北工业大学机械工程学院,天津300131 [2]天津大学精密仪器与光电子工程学院,天津300072

出　　处：《中国惯性技术学报》2023年第8期777-782,789,共7页Journal of Chinese Inertial Technology

基　　金：国家自然科学基金(52175461);河北省自然基金(E2022202136);天津市智能制造专项(20201199)。

摘　　要：针对现有点云分割算法精度差、效率低的问题,提出了一种双目视觉空间定位系统中点云目标融合分割算法。首先,使用随机采样一致性算法剔除场景中的背景平面,保留目标点云;其次,改进P-Linkage聚类分割算法,完成三维点云多目标分割,并使用最小包围盒算法测量目标物体尺寸。实验结果表明,所提算法可以将物体测量误差稳定在3.8%以下,运行效率提高1.5~4.4倍,具有良好的自适应能力,实现了目标物体的精确分割和测量。In response to the problems of poor accuracy and low efficiency of existing point cloud segmentation algorithms,a point cloud target fusion segmentation algorithm in binocular visual spatial positioning system is proposed.Firstly,the random sampling consistency algorithm is used to remove background planes from the scene and preserve the target point clouds.Secondly,the P-Linkage clustering algorithm is improved to achieve multi-objective segmentation of 3D point clouds,and the minimum bounding box algorithm is used to measure the size of the target object.Experimental results show that the proposed algorithm can stabilize the object measurement error below 3.8%,improve running efficiency by 1.5 to 4.4 times and has good adaptive ability to achieves accurate segmentation and measurement of the target object.

关 键 词：工业测量 双目视觉 点云 点云分割 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]