一种结合随机采样一致性与主成分分析的点云配准方法  

A point cloud registration method combining random sample consensus and principal component analysis

作　　者：苏宇 刘海燕[1] 李国勇 SU Yu;LIU Haiyan;LI Guoyong(School of Mechanical and Automotive Engineering,Guangxi University of Science and Technology,Liuzhou 545616,China;Department of Electrical and Mechanical Engineering,Vocational Education Central School of Hechi,Hechi 547000,China)

机构地区：[1]广西科技大学机械与汽车工程学院,广西柳州545616 [2]河池职业教育中心学校机电系,广西河池547000

出　　处：《广西科技大学学报》2022年第4期70-77,共8页Journal of Guangxi University of Science and Technology

基　　金：广西中青年教师基础能力提升项目(2020KY8019);桂教职成专业基地发展项目(GJZC2018-65)资助。

摘　　要：针对迭代最近点算法(ICP)对点云初始位姿有较高要求和随机采样一致性算法(RANSAC)不稳定的问题,提出了一种基于RANSAC和主成分分析(PCA)的点云配准方法。基于局部关键点配准、RANSAC+PCA粗匹配、全局优化精配准的框架,利用固有特征(ISS)关键点检测方法和非最大抑制值处理关键点,通过改进的快速点特征直方图(FPFH)优化建立关键点之间的对应关系组,融合过滤筛选、PCA主方向限制,以降低迭代次数且能利用ICP进行精准配准。本文对飞机点云、道路点云和兔子点云的点云配准数据进行特征点检测方法对比实验,实验表明:在相同的迭代次数条件下,RANSAC+PCA+ICP配准方法比RANSAC+ICP配准方法更具有稳定性,运算时间分别降低7.60%、13.49%、15.64%,并具有较好的配准效果。The iterative closest point(ICP) registration algorithm has higher requirements for the initial pose of point clouds and the lower stability of random sample consensus(RANSAC) registration algorithm. To solve these problems, a point cloud registration based on RANSAC and principal component analysis(PCA) was designed. Based on the framework of local keypoint registration,RANSAC+PCA coarse matching, and global TCP optimization registration, the intrinsic shape signatures(ISS) keypoint detection and non-maximum suppression were used to deal with these keypoints. The correspondence between keypoints was established through improved FPFH optimization. It integrated fusion filtering and PCA main direction limitation to reduce the number of iterations and could use the ICP for precise registration. In this paper, the registration data of aircraft point cloud, road point cloud and rabbit point cloud were compared by feature point detection method.The experiment showed that the RANSAC+PCA+ICP registration method was more stable than the RANSAC+ICP registration method under the same number of iterations;the time was reduced by7.60%, 13.49%, and 15.64% respectively. Therefore, the RANSAC+PCA+ICP registration method is more stable and has better registration effect.

关 键 词：点云配准 随机采样一致性(RANSAC) 主成分分析(PCA) 局部关键点 迭代最近点法(ICP) 

分 类 号：TN911.73[电子电信—通信与信息系统] TP391.41[电子电信—信息与通信工程]