形貌约束的多视角点云分阶配准方法  

A Hierarchical Registration Method of Multiview Point Clouds with Shape Constraints

作　　者：林伟 孙殿柱[1] 李延瑞[2] 沈江华 LIN Wei;SUN Dianzhu;LI Yanrui;SHEN Jianghua(College of Mechanical Engineering, Shandong University of Technology, Zibo, Shandong 255049, China;School of Mechanical Engineering, Xi’an Jiaotong University, Xi’an 710049, China)

机构地区：[1]山东理工大学机械工程学院,山东淄博255049 [2]西安交通大学机械工程学院,西安710049

出　　处：《西安交通大学学报》2020年第6期75-81,共7页Journal of Xi'an Jiaotong University

基　　金：国家自然科学基金资助项目(51575326)。

摘　　要：为提高大数据量多视角点云的配准效率,提出一种基于多分辨率模型的多视角点云分阶配准方法。首先根据平坦形貌约束条件对点云进行递归分割,提取所得割集的核心点作为特征点构造多分辨率模型,然后采用迭代最近点算法基于该模型上层数据求解多视角点云的初始变换矩阵,将其作用于模型后逐级求解下层数据的变换矩阵,最终将复合变换矩阵同步作用于原多视角点云,实现原多视角点云的精确配准。实验结果表明,该分阶配准方法可有效缓解点云单一简化结果导致的配准精度与效率之间的矛盾,在显著降低点云规模的前提下实现原始点云精确配准;当点云规模达106级别时,与加权尺度迭代最近点(WSICP)算法相比,该方法的计算效率提高约2.5倍。A method for hierarchical registration of multiview point clouds based on multiresolution model is proposed to improve the registration efficiency of the multiview point clouds with large data volume.Firstly,the point clouds are recursively segmented according to the constraints of smoothness,and the core points of the subsets are extracted as feature points to construct a multiresolution model.Then,the iterative nearest point algorithm is used to solve the initial transformation matrix of multiview point clouds based on the upper data of the model,which is applied to the model to solve the transformation matrix of the lower data step by step.The complex transformation matrix is synchronously applied to the original multiview point clouds to achieve the accurate registration of the original multiview point clouds.Experimental results show that the hierarchical registration method effectively alleviates the contradiction between registration accuracy and efficiency caused by single simplification of point clouds,and achieves accurate registration of the original point clouds on the premise of significantly reducing the scale of the point clouds.When the scale of point clouds reaches million level,the efficiency of the proposed algorithm is about 2.5 times higher than that of the weighted scaled iterative closest point algorithm.

关 键 词：平坦形貌 递归分割 形貌多分辨率模型 分阶配准 

分 类 号：TP391.7[自动化与计算机技术—计算机应用技术]