THE APPLICATION OF HMMs CONSISTING OF DIFFERENT NUMBER STATES IN CHINESE SIGN LANGUAGE RECOGNITION

Chinese sign language （CSL） is the language used by many deaf people in China. The signer expresses himself by gestures and hand motions. Therefore CSL recognition is a problem of dynamic continuous signal recognition. Most sign language recognition systems use HMMs（hidden Markov models） presently. But the numbers of basic gestures included in different words are variable, so the accuracy will be affected if all models consist of the same number of states. It is difficult to set the exact number of states manually. In this paper, an approach based on dynamic programming is proposed to estimate the number of the states, and a CSL training and recognition system are implemented based on HMMs consisting of different number of states. The result of the experiments shows that the speed and accuracy of CSL recognition are improved.