Study Of Dynamic Deforming Hand Gesture Tracking Under Complex Background

With the rapid development in computer technology, people are forced to improve the interaction ways between human beings and computer in the future. Hand gestures have been widely used and studied in human-computer interaction, for its friendliness and direct viewing in semantic expression.After analyzed the traditional Mean shift algorithm theory and its deficiency, this article proposes a method of linear prediction of real-time hand gesture tracking. Firstly, this method abstracted hand gesture's H component in HSV color space to build Mean shift target model, and then chose skin color sample pixels by GSM. Secondly, to overcome the problem that low tracking acuteness in case of the much too fast hand gestures, this tracking method makes use of linear prediction function to predict hand gesture's beginning position. In experiment, this tracking method showed well abilities.To improve accuracy of hand gesture tracking more, this article proposed another tracking method, which is based on Mean shift of the region grow, to track dynamic deforming hand gesture. This method according to the hand gesture's high clustering property in color space, making use of region grow method to collect sample pixels of hand gesture, using hand gesture as the beginning centre. So that Mean shift computed the real center of hand gesture's sample set. And further more, hand gesture's real center can be computed continuously in video sequence.At the base of the second method, this article proposes a third method, dynamic and deforming hand gesture tracking, which is based on the spatial-feature domain. By the features of color domain and places information in spatial domain, this method calculates the density of the Kernel probability of the union space. And then computes hand gesture's center position in spatial domain. In fact, the third method is new version of the hand gesture tracking based on region grow and Mean shift.In this article, several video sequences of the moving hand gesture and matlab 7.0 Software are recorded in laboratory to prove the feasibility of the three kinds of algorithms. In other aspect, comparing the three kinds of algorithms with traditional mean shift tracking algorithm, we can get the point that the three kinds of algorithms are more excellent than traditional mean shift tracking algorithm.All the work of the article done in laboratory is based on one moving single hand gesture. However, much more work should be done in order that these tracking methods can be used into practice.