融合PCA的改进ICP激光点云配准算法  

Improved ICP Laser Point Cloud Registration Algorithm Based on PCA

作　　者：王太学 江智 江德港 李柏林[2] 郭彩玲[3] WANG Taixue;JIANG Zhi;JIANG Degang;LI Bailin;GUO Cailing(Graduate School of Tangshan,Southwest Jiaotong University,Tangshan,Hebei 063000,China;School of Mechanical Engineering,Southwest Jiaotong University,Chengdu 610031,China;Key Lab of Intelligent Equipment Digital Design and Process Simulation,Tangshan University,Tangshan,Hebei 063000,China)

机构地区：[1]西南交通大学唐山研究生院,河北唐山063000 [2]西南交通大学机械工程学院,成都610031 [3]唐山学院河北省智能装备数字化设计及过程仿真重点实验室,河北唐山063000

出　　处：《遥感信息》2022年第2期70-76,共7页Remote Sensing Information

基　　金：河北省科技厅重点研发计划项目(20327107D);河北省高层次人才项目(A202001095)。

摘　　要：针对迭代最近点(iterative nearest point,ICP)算法进行点云配准过程中需要输入较好的初值和收敛速度较慢的问题,提出了一种融合主成分分析(principal component analysis,PCA)的改进ICP配准算法。首先,基于PCA计算两组点云的主方向,得到两组点云的初始位姿变换;然后,利用主轴基向量的关系校正主轴方向,完成两组点云的大致配准。针对具有部分重叠区域的点云,粗配准后利用KD-tree算法搜索最近点。根据点云重叠部分,提出一种利用有效点对进行ICP迭代求解算法,利用有效点对中欧式距离较大的点对,完成精配准。通过在公开数据集Bunny和实测数据支座点云上的实验表明,改进算法能够为ICP算法提供较好初值,加速具有局部重叠度的点云配准,同时对配准精度也有较高的鲁棒性。In order to solve the problem that the iterative nearest point(ICP)algorithm needs better initial input and has slower convergence speed in point cloud registration process,an improved ICP registration algorithm based on principal component analysis(PCA)is proposed.Firstly,the principal direction of the two groups of point clouds is calculated based on PCA,and the initial pose transformation of the two groups of point clouds is obtained.Then,the principal axis direction is corrected by using the relationship between principal axis basis vectors,and the rough registration of the two groups of point clouds is completed.For the point cloud with partial overlapping region,KD-tree algorithm is used to search the nearest point after rough registration.According to the overlapping part of point cloud,an ICP iterative algorithm using effective point pairs is proposed,which uses the point pairs with large Euclidean distance in the effective point pairs to complete the fine registration.The experimental results on the open data set Bunny and the measured data support point clouds show that the improved algorithm can provide better initial value for ICP algorithm,accelerate the registration of point cloud with local overlap,and also have high robustness to the registration accuracy.

关 键 词：点云配准 主成分分析 主轴校正 ICP 重叠点云 有效点对 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]