文献类型：期刊文献

中文题名：基于保局PCA的三维点云配准算法

英文题名：3Dpoint cloud registration algorithm based on locality preserving PCA

作者：王育坚[1];吴明明[1];高倩[1]

第一作者：王育坚

机构：[1]北京联合大学信息学院

第一机构：北京联合大学智慧城市学院

年份：2018

卷号：44

期号：5

起止页码：562-568

中文期刊名：光学技术

外文期刊名：Optical Technique

收录：CSTPCD;;Scopus;北大核心:【北大核心2017】；CSCD:【CSCD_E2017_2018】；

基金：国家自然科学基金资助项目(61572077)

语种：中文

中文关键词：点云配准;主成分分析;保局投影

外文关键词：point cloud registration;PCA;locality preserving projection

摘要：三维点云配准是三维重建过程中的重要环节,PCA算法应用于点云配准时无法保留点云局部特征,影响了配准效果,故提出一种基于保局PCA的三维点云配准算法。为了保留点云局部特征,采用保局投影LPP的思想,通过K近邻准则构造点云的邻接图及其补图;对邻近点和非邻近点采取不同的处理方式进行特征提取,通过特征矩阵求得转换参数,进行坐标归一化完成配准;为了减少光照噪声影响,对特征向量矩阵前三个主分量加权后求转换参数。实验结果表明,改进算法在对局部特征结构明显的点云进行配准时有较好的效果,改善了对光照噪声的鲁棒性。
Three-dimensional point cloud registration is an important step to reconstruct three dimension model.The local feature of point cloud can not be retained and the registration effect is influenced when the PCA algorithm is applied to point cloud registration.A 3 Dpoint cloud registration algorithm based on locality preserving PCA is proposed.LPP projection is used to preserve the local characteristics of point cloud.LPP constructs the adjacency graph and its complement of point cloud through K-Nearest Neighbor Criterion.The feature extraction is carried out by using different processing methods for adjacent and non-nearest neighbors.The conversion parameters are obtained by the feature matrix,and the coordinates are normalized to complete point cloud registration.The purpose of finding the conversion parameters after weighting the first three principal components of the eigenvector matrix is to reduce the influence of illumination noise.The experimental results show that the improved algorithm has better effect in registration of the point cloud with obvious local feature structure,and the robustness to lighting noise is improved.