基于组合特征点提取与匹配的点云配准研究  

Research on point cloud registration based on combined feature point extraction and matching

作　　者：郑麒麟 罗印升[1] 宋伟[1] ZHENG Qilin;LUO Yinsheng;SONG Wei(Jiangsu University of Technology,Changzhou 213001,China)

机构地区：[1]江苏理工学院,江苏常州213001

出　　处：《无线互联科技》2025年第6期91-95,104,共6页Wireless Internet Science and Technology

基　　金：江苏省科技计划项目,项目编号:BY2022134。

摘　　要：针对点云配准受初始位姿以及冗余数据影响,存在易产生误匹配、配准效率低等问题,文章提出一种基于组合特征点提取与匹配的点云配准方法。首先通过局部协方差矩阵计算特征值偏离比提取的特征点以及超体素质心点组合作为特征点,利用SHOT进行特征描述。然后结合双向最邻近以及改进的随机采样一致性算法筛选匹配关系,实现点云的粗配准。最后使用迭代最近点算法进行优化。实验结果表明,粗配准点云获得了较好位姿,缩短了精配准时间,提高了配准效率,具有较好的配准效果。To address the issues of mismatches and low registration efficiency caused by the influence of initial poses and redundant data in point cloud registration,the article proposes a point cloud registration method based on the extraction and matching of combined feature points.Feature points are selected by combining the feature points extracted using the eigenvalue deviation ratio computed from the local covariance matrix and the supervoxel centroid points.SHOT is used for feature description,and the matching relationships are refined using a combination of the bidirectional nearest neighbor strategy and an improved random sample consensus algorithm.This process achieves coarse registration of point clouds,followed by optimization using the iterative closest point algorithm.Experimental results show that this method provides an accurate initial pose for coarse-registered point clouds,reduces the time required for fine registration,improves registration efficiency,and achieves favorable registration results.

关 键 词：点云配准 特征值偏离比 超体素 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]