局部距离优化的点云配准算法  

Point cloud registration algorithm based on local distance optimization

作　　者：王明[1] 严飞[1,2] WANG Ming;YAN Fei(Automation College,Nanjing University of Information Science and Technology,Nanjing 210044,China;Jiangsu Collaborative Innovation Center of Atmospheric Environment and Equipment Technology,Nanjing 210044,China)

机构地区：[1]南京信息工程大学自动化学院,南京210044 [2]江苏省大气环境与装备技术协同创新中心,南京210044

出　　处：《激光杂志》2023年第2期57-62,共6页Laser Journal

基　　金：江苏省产业前瞻与关键核心技术重点项目(No.BE2020006-2)。

摘　　要：为了降低多视觉点云配准过程中易产生的误匹配等问题,提高配准效率,提出一种加权距离均值的关键点提取算法。以点云表面某点为中心,计算邻近点到中心点处切平面的加权距离均值,以此筛选出具有局部特征信息差异的关键点;选择快速点特征直方图(Point Features Histograms,FPFH)作为关键点的特征描述子;在匹配对应点对方面,采用一种基于邻近匹配对欧氏距离相对一致性的对应关系查找策略,结合随机抽样一致性(Sample Consensus Initial Alignment,RANSAC)算法确定对应点集,得到最优初始变换矩阵完成粗配准;最后使用迭代最近点(Iterative Closest Point,ICP)算法进行精配准。实验结果表明,所提算法在有效提取关键点的基础上提高了对应关系的准确性,较好的初始位姿使得ICP算法最终的收敛速度与传统点云配准算法相比平均提高了约47.75%,具有较好的配准效果。In order to reduce the easy mismatching in the process of multi vision point cloud registration and improve the registration efficiency,this paper proposes a key point extraction algorithm based on weighted distance mean.Taking a point on the surface of the point cloud as the center,the weighted distance mean from the adjacent point to the tangent plane at the central point is calculated to screen the key points with local feature information differences.Fast point feature histogram(FPFH)is selected as the feature descriptor of key points.In the aspect of matching corresponding point pairs,a corresponding relation screening strategy based on the relative consistency of Euclidean distance between adjacent points is designed.Combined with the random sampling consistency(RANSAC)algorithm,the corresponding point set is determined,the optimal initial transformation matrix is obtained,and the rough registration is completed.Finally,the iterative closest point(ICP)algorithm is used to complete the fine registration.The experimental results show that the proposed algorithm improves the accuracy of the corresponding relationship based on the effective extraction of key points.The better initial pose makes the final convergence speed of ICP algorithm increase about 47.75%on average compared with the traditional point cloud registration algorithm,and has better registration results.

关 键 词：机器视觉 点云配准 关键点提取 对应点筛选 迭代最近点 

分 类 号：TN249[电子电信—物理电子学]