Research On Interaction-oriented Gesture Recognition

Along with the gradual development of human civilization, human living condition and computer technology, human-computer interaction is becoming the focus of information industrial competition. It is hoped that in the future the interaction between human and computer can be as natural, accurate and express as that between human themselves. Natural human-computer interaction is the technology that endows computer with such functions as recognizing, understanding and carrying out orders, thoughts and intentions exactly transferred from the interaction in a natural and friendly way.Gesture is a kind of natural and audio-visual model of human-computer interaction. Along with the development and popularization of mobile computer technology, gesture interaction based on wearable equipment has become a research hot point in the field of current human-computer interaction. At present, main stream gesture recognition research usually adopts method of image processing. But since this method is too much dependent on surrounding, people and location, it can hardly make any breakthrough. This thesis makes a study on gesture recognition based on three dimensional accelerometer sensors by combining with such relevant knowledge as natural human-computer interaction model.The main work of this thesis is as follows:First, it introduces relevant knowledge of human-computer interaction, and reviews the present situation of development of gesture research on natural human-computer interaction in both home and abroad.Second, it designs sets of gesture of natural human-computer interaction according to the characteristic of human-computer interaction, and it introduces gesture's collecting plan based on accelerometer sensors.Then, it does an analysis on gesture recognition's effectiveness by decoding data of three dimensional signal collected from main wrist.Finally, the author puts forward an immediate gesture signal detecting and recognizing plan based on accelerometer sensors in terms of gesture signal's unstability, similarity and etc. After gesture accelerometer being auto-detected and preprocessed, a Discrete Hidden Markov Model(DHMM) is built to carry out gesture recognition. In addition to theoretical research, this thesis does analysis on gesture data of both speaker dependent and speaker independent respectively, through experiment, thereby proving the plan designed by the author to be effective to gesture recognition and to be applied extensively.