以运动结构特征为约束条件的序列影像匹配,是基于多变量核密度函数,采用非参数均值漂移方法估计最优局部运动相似性结构特征的过程.核密度函数的带宽大小决定了匹配方法的收敛速度和精度.本文提出了一种可变带宽的自适应运动结构特征的车载全景序列影像匹配方法.首先以采样点在空间域和光流域的局部空间结构定义自适应的带宽矩阵.采用局部光流特征向量的距离加权法,描述光流域上运动相似性结构特征的松弛扩散过程.然后给出自适应多变量核密度函数的表达形式,并探讨了均值漂移向量的求解、终止条件以及种子点的选择方法.最后融合多尺度SIFT描述特征与运动结构特征,建立统一的全景影像匹配框架.试验选择车载移动测量系统获取的城市球全景序列影像,结果表明在内点率变化、物方尺度变化等情况下,本文方法可以实现自适应运动结构特征的相似性度量,提高匹配的正确点数和匹配率,算法表现出较强的稳键性.
Panoramic image matching method with the constraint condition of local structure from motion similarity feature is an important method, the process requires multivariable kernel density estimations for the structure from motion feature used nonparametric mean shift. Proper selection of the kernel bandwidth is a critical step for convergence speed and accuracy of matching method. Variable bandwidth with adaptive structure from motion feature for panoramic image matching method has been proposed in this work. First the bandwidth matrix is defined using the locally adaptive spatial structure of the sampling point in spatial domain and optical flow domain. The relaxation diffusion process of structure from motion similarity feature is described by distance weighting method of local optical flow feature vector. Then the expression form of adaptive multivariate kernel density function is given out, and discusses the solution of the mean shift vector, termination conditions, and the seed point selection method. The final fusions of multi-scale SIFT the features and structure features to establish a unified panoramic image matching framework. The sphere panoramic images from vehicle-borne mobile measurement system are chosen such that a comparison analysis between fixed bandwidth and adaptive bandwidth is carried out in detail. The results show that adaptive bandwidth is good for case with the inlier ratio changes and the object space scale changes. The proposed method can realize the adaptive similarity measure of structure from motion feature, improves the correct matching points and matching rate, experimental results have shown our method to be robust.
[1] MICUSIK B, KOSECKA J. Piecewise Planar City 3D Modeling from Street View Panoramic Sequences[C]//IEEE Conference on Computer Vision and Pattern Recognition. Miami, FL: IEEE, 2009: 2906-2912.
[2] TORII A, HAVLENA M, PAJDLA T. From Google Street View to 3D City Models[C]//IEEE 12th International Conference on Computer Vision Workshops. Kyoto: IEEE, 2009: 2188-2195.
[3] SATO T, YOKOYA N. Multi-baseline Stereo by Maximizing Total Number of Interest Points[C]//Annual Conference SICE. Takamatsu: IEEE, 2007: 1471-1477.
[4] MATSUHISA R, ONO S, KAWASAKI H, et al. Image-based Ego-motion Estimation Using On-vehicle Omnidirectional Camera[J]. International Journal of Intelligent Transportation Systems Research, 2010, 8(2): 106-117.
[5] SATO T, PAJDLA T, YOKOYA N. Epipolar Geometry Estimation for Wide-baseline Omnidirectional Street View Images[C]//IEEE International Conference on Computer Vision Workshops. Barcelona: IEEE, 2011: 56-63.
[6] JI Shunping, SHI Yun. Image Matching and Bundle Adjustment Using Vehicle-based Panoramic Camera[J]. Acta Geodaetica et Cartographica Sinica, 2013, 42(1): 94-100. (季顺平, 史云. 车载全景相机的影像匹配和光束法平差[J]. 测绘学报, 2013, 42(1): 94-100.)
[7] ZHANG Zhengpeng, JIANG Wanshou, ZHANG Jing. A Gross Error Detection Method of Vehicle-borne Cubic Panoramic Image Sequence[J]. Geomatics and Information Science of Wuhan University, 2014, 39(10): 1208-1213. (张正鹏, 江万寿, 张靖. 车载立方体全景影像匹配点的粗差检测方法[J]. 武汉大学学报: 信息科学版, 2014, 39(10): 1208-1213.)
[8] ZHANG Zhengpeng, JIANG Wanshou, ZHANG Jing. An Image Match Method Based on Optical Flow Feature Clustering for Vehicle-borne Panoramic Image Sequence[J]. Acta Geodaetica et Cartographica Sinica, 2014, 43(12): 1266-1273. (张正鹏, 江万寿, 张靖. 光流特征聚类的车载全景序列影像匹配方法[J]. 测绘学报, 2014, 43(12): 1266-1273.)
[9] DEVROYE L, LUGOSI G. Variable Kernel Estimates: on the Impossibility of Tuning the Parameters[C]//GINÉ E, MASON D M, WELLNER J A. High Dimensional Probability II. Birkhäuser Boston: Springer, 2000: 405-424.
[10] BERLINET A, BIAU G, ROUVIèRE L. Optimal L1 Bandwidth Selection for Variable Kernel Density Estimates[J]. Statistics & Probability Letters, 2005, 74(2): 116-128.
[11] TURLACH B A. Bandwidth Selection in Kernel Density Estimation: A Review[M]. [S.l.]: Université Catholique De Louvain, 1993.
[12] ZHOU Fangfang, FAN Xiaoping, YE Zhen. Mean Shift Research and Applications[J]. Control and Decision, 2007, 22(8): 841-847. (周芳芳, 樊晓平, 叶榛. 均值漂移算法的研究与应用[J]. 控制与决策, 2007, 22(8): 841-847.)
[13] ZHOU Jiaxiang, ZHU Jianjun, MEI Xiaoming, et al. An Adaptive Mean Shift Segmentation Method of Remote Sensing Images Based on Multi-dimension Features[J]. Geomatics and Information Science of Wuhan University, 2012, 37(4): 419-422, 440. (周家香, 朱建军, 梅小明, 等. 多维特征自适应MeanShift遥感图像分割方法[J]. 武汉大学学报: 信息科学版, 2012, 37(4): 419-422, 440.)
[14] COMANICIU D. An Algorithm for Data-driven Bandwidth Selection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2003, 25(2): 281-288.
[15] COMANICIU D, RAMESH V, MEER P. The Variable Bandwidth Mean Shift and Data-driven Scale Selection[C]//8th IEEE International Conference on Computer Vision. Vancouver, BC: IEEE, 2001: 438-445.
[16] GILANI S Z A, RAO N I. Data Driven Bandwidth for Medoid Shift Algorithm[C]//MURGANTE B, GERVASI O, IGLESIAS A, et al. Computational Science and Its Applications. Berlin: Springer, 2011: 534-546.
[17] MAYER A, GREENSPAN H. An Adaptive Mean-shift Framework for MRI Brain Segmentation[J]. IEEE Transactions on Medical Imaging, 2009, 28(8): 1238-1250.
[18] COLLINS R T. Mean-shift Blob Tracking through Scale Space[C]//DANIELLE M. IEEE Computer Society Conference on Computer Vision and Pattern Recognition: Vol 2. Baltimore: Victor Graphics, 2003: 234-240.
[19] COMANICIU D, RAMESH V, MEER P. Kernel-based Object Tracking[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2003, 25(5): 564-575.
[20] WANG Jue, THIESSON B, XU Yingqing, et al. Image and Video Segmentation by Anisotropic Kernel Mean Shift[C]//PAJDLA T, MATAS J. Proceedings of the 8th European Conference on Computer Vision. Berlin: Springer, 2004: 238-249.
[21] LOWE D G. Distinctive Image Features from Scale-invariant Keypoints[J]. International Journal of Computer Vision, 2004, 60(2): 91-110.
[22] SUN Deqing, ROTH S, BLACK M J. Secrets of Optical Flow Estimation and Their Principles[C]//IEEE Conference on Computer Vision and Pattern Recognition. San Francisco, CA: IEEE, 2010: 2432-2439.
