The technology of natural Human-Computer Interaction has become an important research content in the field of Computer Application as the Human-Oriented concept proposed.As an important manner of Human-Computer Interaction,gesture recognition is usually based on the method of image processing,though the research based on this approach has acquired some achievements,the dependence on external conditions such as lighting and background limits its popularity,especially in some special occasions and for some special.At the same time,the existing interaction model can not support natural interactive manner such as gestures and voices well.To solve the problem of the limitation in gesture recognition using image-based method,this dissertation presents a method of three dimensional accelerometer based gesture recognition.First of all,a gesture is divided into several atomic gestures,the three dimensional accelerations collected in the process of the gesture implementation are made to be the feature.Then a discrete Hidden Markov Model(HMM)is built for each atomic gesture with the feature of the gestures after the pre-treatment.Using Baum-Welch algorithm to optimize the models while adopting the interpolation method to overcome the lack of training data.Finally,the models recognize atomic gestures which make up to complex and continuous gestures.On the basis of analyzing and referring to the existing interaction models,this dissertation builds a new interaction model which support gestural input and timely feedback,then formalizes the interaction model detailedly.The model is made up of five component,i.e.User,Interaction Object Component,Presentation Component,Dialogue Component and Application Component,under the support of Interaction Object Component,User can interact with the other Components using Manner,Concept and Task Primitives.After theoretical research,this dissertation does some experiments to verify the method of three dimensional accelerometer based gesture recognition.Four Self-defined gestures and the writing of Arabic numerals have been recognized using the method states above.This research not only brings the three dimensional accelerometer based gesture recognition method into Human-Computer Interaction,but also provides a theoretical expansion for gesture recognition and a framework for natural Human-Computer Interaction. |