A Markov Clustering Method for Analyzing Movement Trajectories

In this study we analyze monkeys´ hand movement; our strategy is compositional, division of complex movement into basic simple components-primitives. Representing each trajectory segment as vectors of directions, we model the movement trajectory as a large Markov process where each state is related with an average trajectory pattern. In the next step, in order to find the movements primitives, we cluster the Markov states according to their probabilistic similarity. We present an information theoretic co-clustering algorithm which can be interpreted as a block-matrix approximation of the Markov transition matrix. The performance of the suggested approach is demonstrated on real recorded data.