Scalable Subspace Clustering with Application to Motion Segmentation

Subspace clustering is the problem of grouping an unlabeled set of points into a number of clusters corresponding to subspaces of the ambient space. This problem has applications in unsupervised learning and computer vision. One of the computer vision applications is motion segmentation, where a number of feature point trajectories need to be grouped into a small number of clusters according to their common motion model. The feature point trajectories are obtained by detecting a number of feature points using an interest point

in Bayesian

them through many frames using a feature point tracker or an optical flow algorithm.