A Coarse-to-Fine Approach for Motion Pattern Discovery

Abstract

In this paper, we propose a coarse-to-fine approach to discovery motion patterns. There are two phases in the proposed approach. In the first phase, the proposed median-based GMM achieves coarse clustering. Moreover, the number of clusters can be heuristically found by the proposed algorithm. In the second phase, to refine coarse clustering in the first phase, a Fisher optimal division method is proposed to examine the boundary data points and to detect the change point between motion patterns. The experimental results show that the proposed approach outperforms the existing algorithms.