On The Statistical Model For Sign Language Understanding

The aim of research on sign language recognition is to enable the communication between the hearing impaired and hearing-abled to be freely and to enhance the capacity of body language understanding of computers. As the research history of sign language recognition is only about ten years, most mechanisms for sign language recognition are not much clearer than those for speech recognition. There are many challenges in sign language recognition. They include: how to extract the signer position independent hand feature? How to get the transition model from the end position of one gesture to the start position of the next? What are the smallest recognition units? How to segment these units from continuous gestures? What is the sign language recognition approach with scalability about the vocabularies? Is it possible to recognize sign language effectively only using the information of hand gesture? How to solve problems for the signer independent sign language recognition? How to recognize sign language mixed with hand gesture signal and finger spelling signal? The solutions of the eight problems mentioned above are very important to sign language recognition. Only six preceding problems are investigated in this paper. Based on this, a signer dependent large vocabulary sign language recognition system was designed and implemented.Firstly, the hand feature extraction approach which is independent of signer position is proposed in this paper. The characteristics of the relative space relation of the left hand and right hand are used in this approach.Secondly, as the order of the number of combinations of different states in left hand and right hand is about 10~8, the idea to find recognition unit in each data stream is proposed in this paper, and the multi-stream HMM is also discussed and the multi-stream ANN/HMM is proposed in this paper. The DP is used in encoding the hand gestures.Thirdly, the sign language recognition approach with scalability...