Kinect-based Multi Granularity Research Of Gesture Recognition

Human-computer interaction is an important section in the field of computer science and it is also a cross-functional integration of computer, artificial intelligence, cognitive science and psychology, it has always been a crucial topic of many scholars and research. Gesture is a way of communicating that is essential among people, at present, it attracts more attention than before in the research field of human-computer interaction. Recently, major popular smart phones have finger touch operation systems and some high-end gaming devices also provide gestures control functions. Gestures recognition plays a crucial part in the communicating between people and computers. Traditional gesture can be recognized by camera technology based on optics, and the results of recognition can be significantly impacted by illumination, texture and complex background etc. And simultaneously, an obstacle confronted in research is the variation of degree of freedom of body structure, people actions and accurate express gesture etc.The research work of this dissertation consists of two aspects:1) A research of multi-granularity gesture recognition method in remote cases; According to Kinect skeleton information provided in remote case, analyzed the characteristics of gestures and proposed multi-granularity identification methods, and improve dynamic time warping algorithm in order to provide users to add new gestures function online.2) A research of multi-granularity gesture recognition method in close cases; Due to the close range, Kinect is unable to provide accurate information about the skeleton, in order to achieve the gesture recognition of multi-granularity, follow the depth image of Kinect and extraction of upper body skeleton information, integrated with the algorithm of dynamic time warping to complete the gesture recognition in close cases.