Research On Vision-Based Gesture Recognition And Human-Computer Interaction

The development direction of Human-Computer Interface (HCI) technology has turned from computer oriented to human oriented. Gesture is a nature, direct and easy-learning interface. Using Hand gesture in HCI is a promising technology. As the key of gesture interface, vision based hand gesture recognizing technology has been studied extensively in recent years. This dissertation mainly studies vision based gesture recognition and the application of which in natural HCI.First,this dissertation analyzes and summarizes the existing gesture recognizing technology. Then the key parts of the gesture recognizing technology are introduced with the current stage and development trend.Then important parts of gesture recognizing technology are studied. In the first place, different color spaces are compared by the skin color allocation. HSV color space is chosen for hand gesture segmentation. After that, gesture image is processed by image filtering algorithm. For static gesture recognition, this dissertation studies hand contour extraction algorithm and chooses 8-connected region searching algorithm to extract the hand contour. Extraction of gesture features is studied by analyzing the structure of hand contours. Then a kind of feature extraction algorithm based on both structure and statistics features. In research on dynamic gesture recognition, this dissertation concentrates on gesture tracking algorithm. Camshift tracking algorithm is used to track hand gesture and the tracking algorithm is optimized.To recognize the gesture using extracted feature, a layered recognition algorithm is proposed, which is designed according to the characteristic of the extracted gesture feature. This layered recognition algorithm simplifies the recognition step and enhances the real-time capacity of system using differences between features. To identify the dynamic gesture, a district-divided recognition algorithm is designed according to the requirement of"nature interface".Finally, a real-time gesture recognition system under Microsoft Visual C++6.0 is established, in which 8 statics gestures and 4 dynamic gestures are defined to design the human-computer interface. According to the experiment results, the validity of algorithms proposed in this dissertation is proved.