Research Of Methods Of Tracking And Modeling Of Dynamic Gesture In Image Sequence

With the rapid development and wide application of the computer software and hardware technology, the interaction between human and computer are getting closer. People often get all kinds of information by computer, learning, research, communication and entertainment, promoting the presented and development of the advanced human-computer interaction technology. Hand gesture, the most important interpersonal communication way except the natural language among people, is also introduced into HCI (Human-Computer Interaction), making the interaction between people and computers can be in a more natural and intuitive manner.From the requirement of natural human-computer interaction, this article studied the static hand shape extraction, dynamic hand gesture modeling and feature extraction, as well as hand gesture tracking and recognition in complex background. Finally, based on the simple nature hand gesture, 5 kinds of mouse hand gestures is realized to verify the feasibility of the algorithm.According to the inaccuracy of extract hand shape based on the single clue, a method of hand shape extraction combined color, motion, contour is proposed. It uses H,S and Y component in the HSV and YCbCr color space to realize the skin detection with the illumination change conditions, remove skin background by using the method of frame difference based on motion detection, finally realize the accurate extraction static hand shape by using fusion clues include color, motion, and contour information.In order to overcome the limitation of traditional Mean shift algorithm in complex background and illumination change such as tracking unstable, tracking failure problems can not be restored, we proposed a robust Mean shift hand gesture tracking algorithm fusing hand skin and structural feature. It combined frame difference and skin detection to form targets detection module, which could detect target automatically at the beginning of the tracking procedure and determine the rectangular areas of the target automatically according to tracking result. Compared with traditional Mean shift algorithm, this method improves the accuracy and stability of hand gesture tracking. When the gesture tracking fails caused by fast motion and occlusion, hand tracking can be restored by using object detection. This can improve tracking continuity. Considering the accuracy of gesture modeling is inconsistent with its efficiency, a dynamic hand gesture modeling and real-time extraction method is proposed. It uses starting hand shape, end hand shape and middle trajectory as dynamic gestures model, using boundary moment to extract dynamic hand gesture features. Experiments show it can improve the speed of the dynamic gesture recognition.At last, a simulation experiment which could identify 5 kings of mouse gestures captured from USB camera hava been made in Matlab7.0 programming environment. The experiment results demonstrated that the proposed method is more effective and accurate.