Sign language recognition is a typical application of human computer interaction,sign language is not only the language for deaf to exchange idea,but also is a vital way to communicate with the outside world.The study of sign language recognition can help the deaf communicate with people who are not without obstacles,even further enchance the computer's ability to understand the language of the human gesture.But at present,most of the research on sign language recognition is based on the recognition of sign language vocabulary.because there are some difficulties in the vocabulary segmentation of sign language sentences,so there is little research on the recognition of sign language sentences.To recognize the natural language,split out the word contained in the sentences is the most important step.Therefore,this paper proposes a method of continuous sign language recognition algorithm based on key frame.Key frames can be process as the basic unit of sign language,we can get vocabulary based on key frames,then make the vocabulary into meaningful sentences according to the Hidden Markov Model,realize the recognition of continuous sentences,avoid the difficulty of direct segmentation of sign language.In this paper,firstly,we propose an adaptive key frame extraction algorithm based on sign language locus,and then design a kind of incremental random forest which supports online learning to recognize the key frames.Next we design a sign language sentence recognition algorithm based on weighted key frame,finally this paper design a continuous sign language recognition system for non specific populations,and complete recognized task in the human-computer nature interaction condition.The whole system is only need a somatosensory device and a PC machine,easy to use,no special requirements on the environment.Experiments show that the algorithm of continuous sign language recognition proposed in this paper has high stability and good fault tolerance,it can recognize meaningful sign language sentences in real time. |