在众多的点云配准算法中，ICP算法以其所需的信息少，配准精度高而被广泛使用。然而，因其算法迭代最优化的特点，ICP本身存在时间复杂度高、易受噪声及离群点影响等缺点。针对这些问题，提出一种基于高斯曲率的ICP改进方法。该方法首先利用高斯曲率在刚体变换中保持不变的性质，对配准点云中每个点进行高斯曲率估计；其次，通过设置阈值将配准非关键点及噪声点和离群点滤除；最后，对只包含关键点的点云使用ICP进行配准。实验结果表明，在保证配准精度的前提下，本方法不仅能显著地改善ICP的运行效率，也能有效地提高其抗噪声和离群点的能力。
As one of basic topics of computer vision, 3-D point cloud registration has been studied for decades. Among the works, ICP (iterative closest point) is the most well-known algorithm for its simplicity and accuracy. However, due to its iteratively greedy strategy, ICP is time consuming, prone to local minima, and hence susceptible to noise and outliers. In the work, we present a method to improve the performance of ICP in terms of both efficiency and robustness. Firstly, the method estimates Gaussian curvature of each point in the point clouds. Secondly, the method filters out those points which are considered to be trivial points, outliers, and noise. Then, the method applies ICP to the remaining points. The results demonstrate that our method improves both efficiency and resilience against outliers and noise of ICP without causing accuracy degeneration.
[1] Colas F, Siegwart R. A review of point cloud registration algorithms for mobile robotics[M]. Boston:Now Publishers Inc, 2015:1-104.
[2] Besl P J, Mckay N D. A method for registration of 3-D shapes[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2002, 14(2):239-256.
[3] Rusinkiewicz S, Levoy M. Efficient variants of the ICP algorithm[C]//Proceedings of 3rd International Conference on 3D Digital Imaging and Modeling. Quebec City:IEEE Computer Society, 2002:145-152.
[4] Mavridis P, Andreadis A, Papaioannou G. Efficient sparse ICP[M]. Amsterdam:Elsevier Science Publishers B V, 2015:16-26.
[5] Masuda T, Sakaue K, Yokoya N. Registration and integration of multiple range images for 3-D model construction[C]//Proceedings of International Conference on Pattern Recognition. Vienna:IEEE Computer Society, 1996:879-883.
[6] Godin G. Three-dimensional registration using range and intensity information[C]//International Society for Optics and Photonics. Videometrics Ⅲ. Boston:Videometrics Ⅲ, 1994:279-290.
[7] Weik S. Registration of 3-D partial surface models using luminance and depth information[C]//Proceedings of 3-D Digital Imaging and Modeling. Montreal:IEEE Computer Society, 1997:93-100.
[8] Yang J, Li H, Jia Y. Go-ICP:solving 3D registration efficiently and globally optimally[C]//Proceedings of IEEE International Conference on Computer Vision. Sydney:IEEE Computer Society, 2013:1 457-1 464.
[9] Segal A, Hähnel D, Thrun S. Generalized-ICP[C]//Proceedings of Robotics:Science and Systems V. Seattle:The MIT Press, 2009:Doi:10.15607/RSS.2009.V.021.
[10] Gelfand N, Rusinkiewicz S, Ikemoto L, et al. Geometrically stable sampling for the ICP algorithm[C]//Proceedings of International Conference on 3-D Digital Imaging and Modeling. Banff:IEEE Computer Society, 2003:260-267.
[11] Bosse M, Zlot R. Keypoint design and evaluation for place recognition in 2D lidar maps[J]. Robotics & Autonomous Systems, 2009, 57(12):1 211-1 224.
[12] Magid E, Soldea O, Rivlin E. A comparison of Gaussian and mean curvature estimation methods on triangular meshes of range image data[J]. Computer Vision & Image Understanding, 2007, 107(3):139-159.
[13] Alboul L, Van Damme R. Polyhedral metrics in surface reconstruction:tight triangulations[C]//Proceedings of the 6th IMA Conference Conference on the Mathematics of Surfaces. London:Clarendon Press, 1994:171-200.
[14] Watanabe K, Belyaev A G. Detection of salient curvature features on polygonal surfaces[J]. Computer Graphics Forum, 2001, 20(3):385-392.
[15] Taubin G. Estimating the tensor of curvature of a surface from a polyhedral approximation[C]//Procedings of International Conference on Computer Vision. Cambridge:IEEE Computer Society, 1995:902-907.
[16] Lin C, Perry M J. Shape description using surface triangulation[J]. Proceedings of the IEEE Workshop on Computer Vision, 1982, 26(1):51-65.
[17] Zhang X, Li H, Cheng Z, et al. Robust curvature estimation and geometry analysis of 3D point cloud surfaces[J]. Journal of Information & Computational Science, 2009, 6(5):1 983-1 990.
[18] 马骊溟, 徐毅, 李泽湘. 基于高斯曲率极值点的散乱点云数据特征点提取[J]. 系统仿真学报, 2008, 20(9):2 341-2 344.
[19] 吕震, 柯映林, 孙庆, 等. 反求工程中过渡曲面特征提取算法研究[J]. 计算机集成制造系统, 2003, 9(2):154-157.
[20] Zhang X, Li H, Cheng Z, et al. Curvature estimation of 3d point cloud surfaces through the fitting of normal section curvatures[C]//Proceedings of AsiaGraph. Tokyo, 2008:72-79.
[21] Turk G. Zippered polygon meshes from range images[C]//Proceedings of Conference on Computer Graphics and Interactive Techniques. Orlando:ACM, 1994:311-318.
