Given 2D trajectories of feature points moving in the image frame, this program classifies them into independently moving motions, using the multi-stage unsupervised learning method. This algorithm does segmentation by unsupervised learning of degenerate motions followed by unsupervised learning of general motions. The identity of each point (e.g, 0 for the background and 1 for a moving object) is returned to the standard output.
Yasuyuki Sugaya and Kenichi Kanatani, Multi-stage optimization for multi-body motion segmentation, IEICE Transactions on Information and Systems, Vol. E87-D, No. 7, July 2004, pp. 1935-1942 (.pdf).
Yasuyuki Sugaya and Kenichi Kanatani, Geometric structure of degeneracy for multi-body motion segmentation, D. Comaniciu et al. (Eds.), Statistical Methods in Video Processing, Lecture Notes in Computer Science, No. 3247, Springer-Verlag, Berlin, December 2004, pp. 13-25 (.pdf).
Given 2D trajectories of feature points moving in the image frame, this program detects false trajectories by robustly fitting an appropriate subspace to them and find those that have large residuals. The identity of each point (0 for outliers and 1 for inliers) is indicated in the standard output.
Y. Sugaya and K. Kanatani, Outlier removal for motion tracking by subspace separation, IEICE Transactions on Information and Systems, Vol. E86-D, No. 6 (2003-6), pp. 1095-1102 (.pdf).
Given 2D trajectories of feature points moving in the image frame, this program estimates the number of independent motions using model selection by the geometric AIC (default) or the geometric MDL (option) based on the subspace separation method (default) or the affine space separation method (option).
K. Kanatani and C.Matsunaga, Estimating the number of independent motions for multibody segmentation, Proc. 5th Asian Conf. Computer Vision (ACCV 2002), January, 2002, Melbourne, Australia, Vol. 1, pp. 7-12 (.ps.gz, .pdf).
K. Kanatani, Motion segmentation by subspace separation: Model selection and reliability evaluation Int. J. Image Graphics, Vol. 2, No. 2 (2002-4), 179-197 (.ps.gz, .pdf)
Given 2D trajectories of feature points moving in the image frame, this program classifies them into independently moving motions, using the affine space separation algorithm. This algorithm fits an appropriate affine space to the trajectory data using model selection by the geometric AIC and robust fitting by LMedS. The identity of each point (e.g, 0 for the background and 1 for a moving object) is indicated in the standard output.
K. Kanatani, Evaluation and selection of models for motion segmentation, Proc. 7th Euro. Conf. Computer Vision (ECCV 2002), May 2002, Copenhagen, Denmark, Vol.3, pp. 335-349 (.ps.gz, .pdf).
Given 2D trajectories of feature points moving in the image frame, this program classifies them into independently moving motions, using the subspace separation algorithm. This algorithm fits an appropriate subspace to the trajectory data using model selection by the geometric AIC and robust fitting by LMedS. The identity of each point (e.g, 0 for the background and 1 for a moving object) is indicated in the standard output.
K. Kanatani, Motion segmentation by subspace separation and model selection, Proc. 8th Int. Conf. Computer Vision (ICCV 2001), July, 2001, Vancouver, Canada, Vol. 2, pp. 301-306 (.ps.gz, .pdf).
K. Kanatani, Motion segmentation by subspace separation: Model selection and reliability evaluation Int. J. Image Graphics, Vol. 2, No. 2 (2002-4), 179-197 (.ps.gz, .pdf)