A spectral clustering approach to motion segmentation based on motion trajectory

Multibody motion segmentation is important in many computer vision tasks. This paper presents a novel spectral clustering approach to motion segmentation based on motion trajectory. We introduce a new affinity matrix based on the motion trajectory and map the feature points into a low dimensional subspace. The feature points are clustered in this subspace using a graph spectral approach. By computing the sensitivities of the larger eigenvalues of a related Markov transition matrix with respect to perturbations in affinity matrix, we improve the piecewise constant eigenvectors condition [M. Meila et al., 2001] dramatically. This makes clustering much reliable and robust. We confirm it by experiments.