A sign language recognition system using hidden markov model and context sensitive search

Hand gesture is one of the most natural and expressive ways for the hearing impaired. However, because of the complexity of dynamic gestures, most researches are focused either on static gestures, postures, or a small set of dynamic gestures. As real-time recognition of a large set of dynamic gestures is considered, some efficient algorithms and models are needed. To solve this problem in Taiwanese Sign Language, a statistics based context sensitive model is presented and both gestures and postures can be successfully recognized. A gesture is decomposed as a sequence of postures and the postures can be quickly recognized using hidden Markov model. With the probability resulted from hidden Markov model and the probability of each gesture in a lexicon, a gesture can be easily recognized in a linguistic way in real-time.