面向大部件三维重建的多视角点云配准方法  

Multi-view Point Cloud Registration Method for Large-Scale Components Reconstruction

作　　者：张瑞程 陈坤勇 赵勇[1] ZHANG Ruicheng;CHEN Kunyong;ZHAO Yong(Shanghai Key Laboratory of Digital Manufacture for Thin-walled Structures,Shanghai Jiao Tong University,Shanghai 200240,China)

机构地区：[1]上海交通大学上海市复杂薄板结构数字化制造重点实验室,上海200240

出　　处：《机械设计与研究》2022年第3期30-36,共7页Machine Design And Research

基　　金：国家自然科学基金(51975349);国家重点研发计划(2019YFA0709000)资助项目

摘　　要：针对大尺寸部件三维重建中点云配准易产生错误匹配点和累积误差的问题,提出一种改进的多视角点云配准方法。首先采用基于鲁棒核函数的成对点云配准算法,在点到平面距离度量的ICP算法的基础上通过改进目标函数来减小误匹配点对配准结果的影响。点云对两两顺序配准后,采用基于位姿图优化的多视角点云全局配准方法,将点云全局位姿优化转化为图优化问题求解,通过最小化点云对整体配准误差实现对累积误差的消除。实验结果表明该成对配准方法对点云重叠比值具有较高的鲁棒性,在重叠比值大于40%的点云对上全部配准成功,全局配准方法在顺序配准的基础上将多视角配准误差降低了39.1%,有效减小了顺序配准后的累积误差。Aiming at the problems that the point cloud registration in the 3D reconstruction of large-scale components easily produced mismatched points and accumulated errors,an improved multi-view point cloud registration method is proposed.Based on the point-to-plane iterative closest point algorithm,a paired point cloud registration method based on the robust kernel function is proposed to reduce the influence of mismatched points on the registration result by improving the objective function.After the point cloud pair is sequentially registered,the multi-view point cloud global registration based on pose graph optimization is used to convert point cloud global pose optimization into a graph optimization problem.Pose graph optimization eliminated the sequential registration accumulated errors by minimizing the overall registration error.As is confirmed by experiments,this pairwise registration method has high robustness to point cloud overlap ratio,and all pairs with an overlap ratio greater than 40%are successfully registered.Based on sequential registration,the proposed global registration method reduced the multi-view registration error by 39.1%,which effectively reduced the cumulative error.

关 键 词：成对点云配准 点云全局配准 鲁棒核函数 位姿图优化 

分 类 号：TP391.7[自动化与计算机技术—计算机应用技术]