Hidden Markov Models Based Dynamic Hand Gesture Recog- Nition With Incremental Learning Method

Nowadays, with the development of computer technology, there have sprung up various intelligent interactive methods in the field of human-computer interaction, such as fingerprint, voice or speech, hand gesture and so on. Among these intelligent inter-active methods, hand gesture and hang gesture based applications have occupied a sig-nificant position, and have been maturely applied in the area of robotic control, motion sensing game industry.In the research of dynamic hand gesture recognition, there have already existed various classic hand gesture tracking and corresponding recognition algorithms. In these algorithms, skin color threshold detection, dynamic moving objec tracking meth-ods et al. have been widely applied in hand tracking. As a result of gesturers’ operating habits and the influence of external factors such as lightness, background, gesture’s op-erating environment need some certain limitations. Besides, gesturers should try to avoid face interference during operating to improve the hand tracking accuracy. In terms of hand gesture recognition, due to the advantages of modeling time-series da-ta, the traditional machine learning methods, such as neural network, hidden Markov models et al. emerge from various classical machine learning methods. But once the models in these traditional machine learning methods have been trained in the training phrase, the models will never change in the entire recognition phrase. If the new rec-ognizing gestures have some different changes with the old mode gestures, the system recognition rate could decrease greatly.In order to overcome the shortages of the traditional methods, this paper puts forward an hidden Markov models based dynamic hand gesture recognition with in-cremental learning method, to optimize and improve the original method. These im-provements are:(1) In gesture recognition phrase, the incremental learning methods can dynami-cally adjust the parameters of models during recognition phrase, and improve the sys-tem’s adaptability to new gesture. And expand the traditional three blocks hand gesture recognition system to four blocks in incremental learning environment.(2) In hand gesture tracking phrase, a combination of gesture tracking methods-YCrCb color space threshold skin detection, Adaboost face detection, Camshift mov-ing hand tracking-can efficiently track dynamic hand gestures in a normal operating environment, and also effectively remove the interference of face.(3) In the introduction of hidden Markov models chapter, we amply explain the details of algorithms, the topological structure of models, scale factor, principles of multi-samples training and the differences between recognized algorithms in recogni-tion phrase.Finally, the experiment results show that comparing with traditional methods, the proposed system can obtain better robustness and has higher recognition rate for the unknown new gestures.