自适应Harris角点提取的点云粗配准算法  

Coarse Point Cloud Registration Based on Adaptive Harris Corner Extraction

作　　者：王丞 田暄 郭瑞 张玉龙[1] WANG Cheng;TIAN Xuan;GUO Rui;ZHANG Yulong(School of Software Engineering, Xi’an Jiaotong University, Xi’an 710049, China;State Key Laboratory of Rail Transit Engineering Informatization (FSDI), Xi’an 710043, China)

机构地区：[1]西安交通大学软件学院,西安710049 [2]轨道交通工程信息化国家重点实验室(中铁一院),西安710043

出　　处：《西安交通大学学报》2022年第3期33-44,共12页Journal of Xi'an Jiaotong University

基　　金：轨道交通工程信息化国家重点实验室(中铁一院)开放研究课题资助项目(SKLK16-09)。

摘　　要：针对传统3D-Harris角点提取算法中,Harris算子使用降维后的缺失几何信息、角点提取时响应值计算量大且耗时长、特征点对匹配精度不高以及需要手动设定角点响应阈值等问题,提出了一种完整而高效的Harris角点自适应特征描述、提取和匹配的点云粗配准算法。引入正交梯度算子对传统Harris算子和自相关函数进行改进;利用点云曲率约束实现角点的自适应筛选与提取,减少角点响应值的计算量;构建角点几何结构的特征描述子,结合阈值检测和描述子匹配,将角点匹配对集合进行扩展,从而完成源点云和目标点云之间粗配准;将所提算法得到的配准结果作为精配准初始值,利用迭代最近点算法实现精配准。与对比算法在公开数据集上进行实验比较,结果表明:所提算法的特征正确提取率为0.93,正确率最高;提取时间为7.63 s,效率最快;所提算法结合精配准步骤在实验数据集上的旋转误差、平移误差和运行时间均为最低,配准效果最佳。Aiming at the problems of the Harris operator using missing geometric information after dimension reducing,large response value calculation,huge time consumption,low matching accuracy of feature point pairs,and setting the corner response threshold manually,we propose a complete and efficient adaptive coarse point cloud registration method for corner description,extraction and matching.Firstly,the orthogonal gradient operator is introduced to improve the traditional Harris operator.Then,the point cloud curvature constraint is used to realize the adaptive screening and extraction of corner points,so as to make full use of the geometric information of the point cloud,reduce the calculation amount of the corner point response value,and improve the accuracy and efficiency of corner point extraction.Secondly,the feature descriptor based on intrinsic shape signatures is proposed to construct the corner geometric structure.Then combined with threshold detection and descriptor matching,the corner point matching is expanded to the set,so as to complete the coarse registration between the source point cloud and the target point cloud.Finally,the iterative closest point algorithm is used to realize the fine registration.The experimental results on the public data set show that:compared with five existing feature extraction and coarse point cloud registration algorithms,the proposed method can effectively mine more point cloud’s local and global features,and obtain the best results.The correct corner matching rate of the proposed algorithm is 0.93,and the extraction time is 7.63 seconds.The proposed method combined with the fine registration step has the lowest rotation error,translation error and algorithm running time on the test data sets.

关 键 词：点云粗配准 HARRIS算子 角点响应 曲率约束 特征描述子 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]