以单目观测下三维刚体目标的姿态估计为研究对象,针对现有迭代估计方法存在的收敛半径小和收敛速度慢的问题,提出一种新的基于2D-3D泛轮廓点对应的迭代姿态估计方法.与现有的基于数值优化的方法不同,本方法从输入图像的2D泛轮廓点出发,着眼于显性地建立输入图像到目标三维模型的2D-3D特征投影对应关系,进而以此显性投影对应关系对目标的三维姿态参数进行估计.实验结果表明,该方法在算法复杂性、收敛半径和收敛速度上均有明显改进.
Pose estimation of 3D rigid object based on monocular vision is studied. Considering the insufficiency of state-of-art iterative methods in convergence radius and convergence speed, we propose a new iterative pose estimation method based on general 2D-3D contour point correspondence. The proposed method focuses on explicitly establishing the 2D-3D feature projection correspondence between the input image and object's 3D model with the general 2D contour points of the input image, and then estimates object's 3D pose parameters. Experiment results show that the proposed method provides notable improvements in algorithm complexity, convergence radius, and convergence speed.
[1] Lu C P, Hager G D, Mjolsness E. Fast and globally convergent pose estimation from video images [J]. IEEE Trans on Pattern Analysis and Machine Intelligence, 2000, 22(6): 610-622.
[2] Iwashita Y, Kurazume R, Konishi K, et al. Fast alignment of 3D geometrical models and 2D grayscale images using 2D distance maps [J]. Systems and Computers in Japan, 2007, 38(4): 1889-1899.
[3] Dambreville S, Sandhu R, Yezzi A, et al. A geometric approach to joint 2D region-based segmentation and 3D pose estimation using a 3D shape prior [J]. SIAM Journal on Imaging Sciences, 2010, 3(1): 110-132.
[4] Haralick R M, Lee C, Ottenberg K, et al. Review and analysis of solution of the three point perspective pose estimation problem [J]. International Journal of Computer Vision, 1994, 13(3): 331-356.
[5] Noguer F M, Lepetit V, Fua P. Accurate non-iterative O(n) solution to the PnP problem [C]//Proceedings of International Conference on Computer Vision, IEEE. 2007: 1-8.
[6] Leng D W, Sun W D. Finding all the solutions of PnP problem [C]//Proceedings of International Workshop on Imaging Systems and Techniques, IEEE. 2009: 348-352.
[7] Dunker J, Hartmann G, Stöhr M. Single view recognition and pose estimation of 3D objects using sets of prototypical views and spatially tolerant contour representations [C]//Proceedings of International Conference on Pattern Recognition. 1996: 14-18.
[8] Hartley R, Zisserman A. Multiple view geometry in computer vision [M]. 2nd ed. United Kingdom: Cambridge University Press, 2003.
[9] Felzenszwalb P F, Huttenlocher D P. Distance transforms of sampled functions [R]. Cornell Computing and Information Science TR2004-1963, 2004.
[10] Fabbri R, Costa L D F, Torelli J C, et al. 2D Euclidean distance transform algorithms: a comparative survey [J]. ACM Computing Surveys, 2008, 40(1): 1-44.
[11] Osher S, Fedkiw R. Level set methods and dynamic implicit surfaces [M]. Springer, 2003.
[12] Struik D J. Lectures on classical differential geometry [M]. 2nd ed. New York: Dover Publications, Inc, 1961.
[13] Smith S M, Brady J M. SUSAN—a new approach to low level image processing [J]. International Journal of Computer Vision, 1997, 23(1): 45-78.
[14] Bertsekas D P. Constrained optimization and Lagrange multiplier methods [M]. Boston: Academic Press, 1982.
