多特征提取与匹配矩阵驱动的点云配准  

Multi feature extraction and matching matrix driven point cloud registration

作　　者：刘磊[1,2] 熊风光 尹宇慧[1,2] 郭锐 薛红新 韩燮[1,2] LIU Lei;XIONG Feng-guang;YIN Yu-hui;GUO Rui;XUE Hong-xin;HAN Xie(School of Data Science and Technology,North University of China,Taiyuan 030051,China;Engineering Research Center for Vision Information Processing and Intelligent Robot of Shanxi Province,Shanxi Development and Reform Commission,Taiyuan 030051,China)

机构地区：[1]中北大学计算机科学与技术学院,山西太原030051 [2]山西省发展和改革委员会,山西省视觉信息处理及智能机器人工程研究中心,山西太原030051

出　　处：《计算机工程与设计》2023年第5期1419-1426,共8页Computer Engineering and Design

基　　金：国家自然科学基金项目(62106238);山西省自然科学基金项目(201901D111150);山西省自然科学基金项目(201901D111149);山西省回国留学人员科研基金项目(2020-113);山西省科技成果转化引导专项基金项目(202104021301055)。

摘　　要：针对三维点云配准时容易受到噪声、奇异值等不利因素影响的问题,提出一种多特征提取与匹配矩阵驱动的点云配准方法。利用多特征提取模块提取稳健的关键点和关键点的局部几何特征;分别计算出关键点的特征匹配矩阵和空间匹配矩阵,并对二者进行融合,提高正确匹配点对的概率;在前两方法的驱动下,利用加权奇异值分解计算点云间的刚体变换矩阵,降低噪声和奇异值等因素对配准的影响。实验结果表明,所提方法与现有的点云配准方法相比,具有更高的配准精度和较强的鲁棒性。Aiming at the fact that 3D point cloud registration is easily affected by unfavorable factors such as noise and singular values,a point cloud registration method driven by multi-feature extraction and matching matrix was proposed.The multi-feature extraction module was used to extract robust key points and local geometric features of key points.The feature matching matrix and spatial matching matrix of key points were calculated respectively,and the two were fused to improve the probability of correct matching point pairs.Driven by the first two methods,the weighted singular value decomposition was used to calculate the rigid transformation matrix between point clouds to reduce the influence of noise and singular values on the registration.Experimental results show that the proposed method has higher registration accuracy and stronger robustness than existing point cloud registration methods.

关 键 词：三维点云 多特征提取 局部几何特征 匹配矩阵 匹配点对 加权奇异值分解 刚体变换 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]