基于3D形状上下文特征改进的ICP点云配准算法  

Improved ICP point cloud registration algorithm based on 3D shape context features

作　　者：朱乾荣 柏艳红[1] 王银[1] 孙志毅[1] 赵兵洋 ZHU Qianrong;BAI Yanhong;WANG Yin;SUN Zhiyi;ZHAO Bingyang(School of Electronic Information Engineering,Taiyuan University of Science and Technology,Taiyuan 030024 China)

机构地区：[1]太原科技大学电子信息工程学院,山西太原030024

出　　处：《海洋测绘》2023年第2期69-73,共5页Hydrographic Surveying and Charting

基　　金：山西省重点研发计划(201903D121130);山西省研究生优秀创新项目(2021Y684)。

摘　　要：针对传统ICP算法配准过程中存在误差大、耗时长及收敛慢等问题,提出一种基于曲率特征约束的3D-Harris算法结合3D形状上下文(3DSC)特征的点云配准方法。首先提取点云数据的体素中心,利用k最近邻搜索获取体素中心邻近点为体素栅格来下采样,以表面法线与平均曲率为特征约束提取降采样后3D-Harris特征点,采用3DSC对特征点进行描述,然后根据3DSC特征结合RANSAC算法进行粗配准,估算点云的初始位姿。最后利用k-d tree加速对应点的查找,线性最小二乘法优化点到面ICP算法求解最优变换矩阵。采用不同规模、不同重叠度且含高斯噪声的缺失数据的数据集进行仿真实验,结果表明:与经典ICP方法及结合SIFT的SAC-IA+ICP方法比较,所提方法在保持算法快速性的情况下,配准精度更高,收敛速度更快。Aiming at the problems of large error,time-consuming and slow convergence in the registration process of the traditional ICP algorithm,a point cloud registration method based on the 3D-Harris algorithm with curvature feature constraints and 3D shape context(3DSC)features is proposed.Firstly,the voxel center of the point cloud data is extracted,and the k nearest neighbor search is used to obtain the adjacent points of the voxel center as the voxel grid for down sampling.Taking the surface normal and average curvature as feature constraints,the 3D-Harris feature points after down sampling are extracted,and 3DSC is used to describe the feature points.Then,according to the 3DSC features combined with the RANSAC algorithm,rough registration is performed to estimate the initial pose of the point cloud.Finally,the k-d tree is used to speed up the search of corresponding points,and the linear least squares method optimizes the point-to-surface ICP algorithm to solve the optimal transformation matrix.Simulation experiments are carried out on datasets with different scales,different overlapping degrees and missing data with Gaussian noise.The results show that,compared with the classical ICP method and the SAC-IA+ICP method combined with SIFT,the proposed method can maintain the rapidity of the algorithm,the registration accuracy is higher and the convergence speed is faster.

关 键 词：点云配准 精配准 点到平面ICP 3D-Harris算法 3D形状上下文 

分 类 号：P236[天文地球—摄影测量与遥感]