融合颜色与几何信息的点云配准  

Point Cloud Registration with Color and Geometric Information

作　　者：陆军[1] 刘霜 乔鹏飞 鲁林超 LU Jun;LIU Shuang;QIAO Pengfei;LU Linchao(College of Intelligent Systems Science and Engineering,Harbin Engineering University,Harbin,Heilongjiang 150001,China)

机构地区：[1]哈尔滨工程大学智能科学与工程学院,黑龙江哈尔滨150001

出　　处：《北京理工大学学报》2024年第4期428-438,共11页Transactions of Beijing Institute of Technology

基　　金：黑龙江省自然科学基金资助项目(F201123)。

摘　　要：目前,大部分点云配准算法是基于点云数据的几何特征进行描述.随着能够同时采集对象坐标和颜色的扫描设备出现,为更好利用颜色信息,对彩色点云中的颜色信息描述进行研究,提出一种基于颜色分布的3DLGOP特征描述子,并将其与几何特征描述子FPFH、颜色特征描述子CSHOT融合,设计出FPFH-3DLGOP的混合描述符,采用最近邻比值法得到初始对应关系,采用随机采样一致性去除错误对应关系,对匹配关系使用奇异值分解(SVD)求得三维刚体变换矩阵,进而完成点云配准.实验表明,所提出的特征描述符充分地利用了点云数据的颜色特征与几何特征,不仅可以很好地完成彩色点云的配准,而且还提高了配准的匹配率和精度.At present,most point cloud registration algorithms are based on the geometric features of point cloud data for description.With the emergence of scanning devices that can simultaneously collect object coordinates and colors,in order to better utilize color information and study the description of color information in color point clouds,a 3DLGOP feature descriptor was proposed based on color distribution,fused with geometric feature descriptors FPFH and color feature descriptors CSHOT to design a mixed descriptor of FPFH-3DLGOP.Using a nearest neighbor ratio method to obtain the initial correspondence,random sampling consistency to removes the wrong correspondence and singular value decomposition(SVD)to obtain the 3D rigid body transformation matrix,a point cloud registration was completed for the matching relationship.The experiment results show that,utilizing the color and geometric features of point cloud data,the proposed feature descriptor can not only complete the registration of color point clouds fully,but also improve the matching rate and accuracy of registration.

关 键 词：点云配准 几何特征 颜色特征 特征融合 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]