A classification method of motion database using hidden Markov model

This paper proposes a classification method of a stored motion-data. Robotic technology has made progress, and robots are demanded to cooperate with human. To realize the human and robot exist together, a motion recognition system is needed. In the conventional method, the stored motion-data is classified in advance to search the motion quickly and accurately. However, the task of the classification will be very complex when the stored data is increased. Therefore, the classification system of stored data automatically is required. Since the human motion is time series information and unsteady signal, a hidden Markov Model is used as the probability models. Additionally, this paper shows that Kullback-Leiblaer divergence indicates the similarity index of the stored motion. At this time, the motion is classified according to the acceleration information, which includes the pure force and position information. The validity of the proposed method is confirmed by simulations.