一种利用关键点提取的改进ICP三维彩色点云场景配准方法  

Improved ICP Registration Method Using Key Point Extraction for Three-dimensional Color Point Cloud

作　　者：杨永涛[1,2] 黄国言[2] 吴培良[2] 

机构地区：[1]燕山大学信息技术中心,河北秦皇岛066004 [2]燕山大学信息科学与工程学院,河北秦皇岛066004

出　　处：《小型微型计算机系统》2017年第9期2139-2145,共7页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61305113)资助;河北省自然科学基金项目(2016203358)资助

摘　　要：为了解决传统ICP算法对初始值的敏感以及配准效率不高的问题,提出一种基于关键点提取的三维彩色点云场景配准新方法.该方法充分利用RGB-D数据所提供的三维彩色点云的有效信息,首先通过特征提取算法检测三维场景的彩色信息关键点,获得相应关键点描述子向量,通过最邻近点算法和向量内积最大值原则,对关键点进行匹配和优化,并对应在三维点云中.在此基础上,利用关键点云的点对曲率一致性分析,对错误匹配的关键点对进行剔除,获得配准率较高的关键点云集.最后利用基于关键点提取的改进ICP配准方法对关键点对进行配准,求取变换矩阵,利用变换矩阵将全部点云数据配准.实验结果表明,本文算法避免了对初始值的敏感以及噪点对配准的影响,在保证配准精度和配准效果的前提下,极大简化了配准点集,大大减少了所用的时间,明显提高了效率,对实际应用具有积极的意义.Traditional ICP algorithm is sensitive to the initial value and has a low registration efficiency,in order to solve these problems,this paper presents a new method for 3D color point cloud scene registration,which is based on the key point extraction.Firstly,this method makes good use of the effective information of 3D color point cloud calculated from RGB-D data,using feature extraction algorithm to detect the key points of the 3D color information,determine the descriptor vector of the corresponding key points,the corresponding matching key points were found and optimized by using the interative closest point algorithm and the maximum principle of vector Inner-product,and converted to 3D point cloud.Then,calculates the key point curvature,and deletes the error matched points by the rule of curvature consistency,to obtain the key points with higher registration efficiency.Finally,calculates the transformation matrix using the improved ICP algorithm,and then registers all the point cloud data.Experimental results show that this method is neither sensitive to the initial value nor influenced by noise points,on the premise of insuring the registration precision,it can greatly simplify the match points,reduce the time overheads,and achieve higher efficiency,the method in this paper is positive to practical applications. 

关 键 词：关键点 曲率一致性 彩色信息 深度图像 描述子向量 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]