Gesture Recognition Using the Linear Combination of Membership Degrees of Observations

This paper introduces a novel gesture recognition method. In the method, hand trajectory is represented by the sequence of symbols and each symbol has a specific membership degree obtained from the genetic algorithm training. In order to determine the membership degree of input observations sequence in a class, the system uses the linear combination of membership degrees of observations in sequence. Because of using negative and positive samples for training gesture classes in the proposed method, the recognition system has a good performance in distinguishing very similar gestures. Experiments show that the method developed in this study outperforms HMM and SOMM methods in different gesture datasets.