几何刚性和法向量采样一致性的点云配准算法
|
摘要
针对三维激光点云配准中随机采样一致性(RANSAC)算法存在采样次数多、准确度低的缺点,该文提出了一种结合几何刚性和法向量一致性的点云配准算法。该算法通过改进采样策略降低采样次数并提升匹配精度,首先将点云的法向量信息加入采样集,使得每次的采样点从三对减少为两对;接着以两对采样点的刚性和法向量一致性计算置信度来确定当前采样是否置信;最后以迭代运算选取采样内点数最高的样本来估算变换矩阵实现点云精确配准。对激光三维点云进行配准试验,结果表明,本文方法在匹配效率及匹配性能上均优于传统RANSAC算法,且配准精度更高。
Considering the large amount of false matches and low efficiency problems exist in the feature matching stage of 3 Dpoint cloud registration methods,this paper proposes a sample consensus algorithm based on rigidity and normal consistency.Traditional random sample consensus(RANSAC)suffers from the requirement of huge iterations and low precision problems,we solve these problems by improving the sampling strategy,first,we consider normal information during sampling so as to reduce the sample size from 3 to 2;second,we test the rigidity and normal consistency of current sampled matches to judge if they are convincing;finally,samples yield to the largest number of inliers are selected for transformation estimation in our iterative algorithm.Experiments on LiDAR point clouds show that the proposed algorithm exceeds RANSAC in terms of both precision and efficiency.
Considering the large amount of false matches and low efficiency problems exist in the feature matching stage of 3 Dpoint cloud registration methods,this paper proposes a sample consensus algorithm based on rigidity and normal consistency.Traditional random sample consensus(RANSAC)suffers from the requirement of huge iterations and low precision problems,we solve these problems by improving the sampling strategy,first,we consider normal information during sampling so as to reduce the sample size from 3 to 2;second,we test the rigidity and normal consistency of current sampled matches to judge if they are convincing;finally,samples yield to the largest number of inliers are selected for transformation estimation in our iterative algorithm.Experiments on LiDAR point clouds show that the proposed algorithm exceeds RANSAC in terms of both precision and efficiency.
引文
[1]SHAH S A A,BENNAMOUN M,BOUSSAID F.A novel3Dvorticity based approach for automatic registration of low resolution range images[J].Pattern Recognition,2015,48(9):2859-2871.
[2]赵煦.基于地面激光扫描点云数据的三维重建方法研究[D].武汉:武汉大学,2010.(ZHAO Xu.Research on 3Dreconstruction of point cloud from terrestrial laser scanning[D].Wuhan:Wuhan University,2010.)
[3]GRUEN A,AKCA D.Least squares 3Dsurface and curve matching[J].ISPRS Journal of Photogrammetry and Remote Sensing,2005,59(3):151-174.
[4]陶海跻,达飞鹏.一种基于法向量的点云自动配准方法[J].中国激光,2013,40(8):179-184.(TAO Haiji,DA Feipeng.Automatic registration algorithm for the point clouds based on the normal vector[J].Chinese Journal of Lasers,2013,40(8):179-184.)
[5]邢正全,邓喀中,薛继群.基于K-近邻搜索的点云初始配准[J].测绘科学,2013,38(2):93-95.(XINGZhengquan,DENG Kazhong,XUE Jiqun.Initial registration for point cloud based on K-nearest neighbor search[J].Science of Surveying and Mapping,2013,38(2):93-95.)
[6]李彩林,郭宝云,季铮.多视角三维激光点云全局优化整体配准算法[J].测绘学报,2015,44(2):183-189.(LI Cailin,GUO Baoyun,JI Zheng.Global optimization and whole registration algorithm of multi-view 3Dlaser point cloud[J].Acta Geodaetica et Cartographica Sinica,2015,44(2):183-189.)
[7]刘斌,郭际明,邓祥祥.结合八叉树和最近点迭代算法的点云配准[J].测绘科学,2016,41(2):130-132.(LIUBin,GUO Jiming,DENG Xiangxiang.A point cloud registration method based on octree and ICP[J].Science of Surveying and Mapping,2016,41(2):130-132.)
[8]CHUA C S,JARVIS R.Point signatures:a new representation for 3dobject recognition[J].International Journal of Computer Vision,1997,25(1):63-85.
[9]RUSU R B,BLODOW N,BEETZ M.Fast point feature histograms(FPFH)for 3Dregistration[C]∥Robotics and Automation,ICRA’09.[S.l.]:IEEE,2009:3212-3217.
[10]GUO Y,SOHEL F,BENNAMOUN M,et al.Rotational projection statistics for 3Dlocal surface description and object recognition[J].International Journal of Computer Vision,2013,105(1):63-86.
[11]YANG J,CAO Z,ZHANG Q.A fast and robust local descriptor for 3Dpoint cloud registration[J].Information Sciences,2016,346:163-179.
[12]YANG J,ZHANG Q,XIAN K,et al.Rotational contour signatures for both real-valued and binary feature representations of 3Dlocal shape[J].Computer Vision and Image Understanding,2017,160:133-147.
[13]袁红星,吴少群,朱仁祥,等.平面单应性矩阵求解的CUDA并行实现[J].微型机与应用,2012,31(23):38-41.(YUAN Hongxing,WU Shaoqun,ZHU Renxiang,et al.CUDA-based parallel implementation of homography estimation[J].Microcomputer&Its Applications,2012,31(23):38-41.)
[14]LEE D,KIM H,MYUNG H.GPU-based real-time RGB-d 3Dslam[C]∥Ubiquitous Robots and Ambient Intelligence(URAI),2012 9th International Conference on.[S.l.]:IEEE,2012:46-48.
[15]FIJANY A,HOSSEINI F.Image processing applications on a low power highly parallel SIMD architecture[C]∥Aerospace Conference.[S.l.]:IEEE,2011:1-12.
[16]伍梦琦,李中伟,钟凯,等.基于几何特征和图像特征的点云自适应拼接方法[J].光学学报,2015,35(2):237-244.(WU Mengqi,LI Zhongwei,ZHONG Kai,et al.Adaptive point cloud registration method based on geometric features and photometric features[J].Acta Optica Sinica,2015,35(2):237-244.)
[17]罗文超,刘国栋,杨海燕.SIFT和改进的RANSAC算法在图像配准中的应用[J].计算机工程与应用,2013,49(15):147-149.(LUO Wenchao,LIU Guodong,YANG Haiyan.Application of SIFT and advanced RANSAC algorithm on image registration[J].Computer Engineering and Applications,2013,49(15):147-149.
[18]佘建国,徐仁桐,陈宁.基于ORB和改进RANSAC算法的图像拼接技术[J].江苏科技大学学报(自然科学版),2015,29(2):164-169.(SHE Jianguo,XURentong,CHEN Ning.Image stitching technology based on ORB and improved RANSAC algorithm[J].Journal of Jiangsu University of Science and Technology(Natural Science Edition),2015,29(2):164-169.)
[19]FISCHLER M A,BOLLES R C.Random sample consensus:aparadigm for model fitting with applications to image analysis and automated cartography[J].Communication of ACM,1981,24(6):381-395.
[20]CHEN H,BHANU B.3Dfree-form object recognition in range images using local surface patches[J].Pattern Recognition Letters,2007,28(10):1252-1262.
[2]赵煦.基于地面激光扫描点云数据的三维重建方法研究[D].武汉:武汉大学,2010.(ZHAO Xu.Research on 3Dreconstruction of point cloud from terrestrial laser scanning[D].Wuhan:Wuhan University,2010.)
[3]GRUEN A,AKCA D.Least squares 3Dsurface and curve matching[J].ISPRS Journal of Photogrammetry and Remote Sensing,2005,59(3):151-174.
[4]陶海跻,达飞鹏.一种基于法向量的点云自动配准方法[J].中国激光,2013,40(8):179-184.(TAO Haiji,DA Feipeng.Automatic registration algorithm for the point clouds based on the normal vector[J].Chinese Journal of Lasers,2013,40(8):179-184.)
[5]邢正全,邓喀中,薛继群.基于K-近邻搜索的点云初始配准[J].测绘科学,2013,38(2):93-95.(XINGZhengquan,DENG Kazhong,XUE Jiqun.Initial registration for point cloud based on K-nearest neighbor search[J].Science of Surveying and Mapping,2013,38(2):93-95.)
[6]李彩林,郭宝云,季铮.多视角三维激光点云全局优化整体配准算法[J].测绘学报,2015,44(2):183-189.(LI Cailin,GUO Baoyun,JI Zheng.Global optimization and whole registration algorithm of multi-view 3Dlaser point cloud[J].Acta Geodaetica et Cartographica Sinica,2015,44(2):183-189.)
[7]刘斌,郭际明,邓祥祥.结合八叉树和最近点迭代算法的点云配准[J].测绘科学,2016,41(2):130-132.(LIUBin,GUO Jiming,DENG Xiangxiang.A point cloud registration method based on octree and ICP[J].Science of Surveying and Mapping,2016,41(2):130-132.)
[8]CHUA C S,JARVIS R.Point signatures:a new representation for 3dobject recognition[J].International Journal of Computer Vision,1997,25(1):63-85.
[9]RUSU R B,BLODOW N,BEETZ M.Fast point feature histograms(FPFH)for 3Dregistration[C]∥Robotics and Automation,ICRA’09.[S.l.]:IEEE,2009:3212-3217.
[10]GUO Y,SOHEL F,BENNAMOUN M,et al.Rotational projection statistics for 3Dlocal surface description and object recognition[J].International Journal of Computer Vision,2013,105(1):63-86.
[11]YANG J,CAO Z,ZHANG Q.A fast and robust local descriptor for 3Dpoint cloud registration[J].Information Sciences,2016,346:163-179.
[12]YANG J,ZHANG Q,XIAN K,et al.Rotational contour signatures for both real-valued and binary feature representations of 3Dlocal shape[J].Computer Vision and Image Understanding,2017,160:133-147.
[13]袁红星,吴少群,朱仁祥,等.平面单应性矩阵求解的CUDA并行实现[J].微型机与应用,2012,31(23):38-41.(YUAN Hongxing,WU Shaoqun,ZHU Renxiang,et al.CUDA-based parallel implementation of homography estimation[J].Microcomputer&Its Applications,2012,31(23):38-41.)
[14]LEE D,KIM H,MYUNG H.GPU-based real-time RGB-d 3Dslam[C]∥Ubiquitous Robots and Ambient Intelligence(URAI),2012 9th International Conference on.[S.l.]:IEEE,2012:46-48.
[15]FIJANY A,HOSSEINI F.Image processing applications on a low power highly parallel SIMD architecture[C]∥Aerospace Conference.[S.l.]:IEEE,2011:1-12.
[16]伍梦琦,李中伟,钟凯,等.基于几何特征和图像特征的点云自适应拼接方法[J].光学学报,2015,35(2):237-244.(WU Mengqi,LI Zhongwei,ZHONG Kai,et al.Adaptive point cloud registration method based on geometric features and photometric features[J].Acta Optica Sinica,2015,35(2):237-244.)
[17]罗文超,刘国栋,杨海燕.SIFT和改进的RANSAC算法在图像配准中的应用[J].计算机工程与应用,2013,49(15):147-149.(LUO Wenchao,LIU Guodong,YANG Haiyan.Application of SIFT and advanced RANSAC algorithm on image registration[J].Computer Engineering and Applications,2013,49(15):147-149.
[18]佘建国,徐仁桐,陈宁.基于ORB和改进RANSAC算法的图像拼接技术[J].江苏科技大学学报(自然科学版),2015,29(2):164-169.(SHE Jianguo,XURentong,CHEN Ning.Image stitching technology based on ORB and improved RANSAC algorithm[J].Journal of Jiangsu University of Science and Technology(Natural Science Edition),2015,29(2):164-169.)
[19]FISCHLER M A,BOLLES R C.Random sample consensus:aparadigm for model fitting with applications to image analysis and automated cartography[J].Communication of ACM,1981,24(6):381-395.
[20]CHEN H,BHANU B.3Dfree-form object recognition in range images using local surface patches[J].Pattern Recognition Letters,2007,28(10):1252-1262.