Research On Algorithm And Model Of Hand Gestures Recognition Based On Computer Vision

Vision-based hand gesture recognition is an approach to get category and semantics of hand gesture with technology of pattern recognition from image or video. The hand gesture is diversity, ambiguity and deformation. Vision-based hand gesture recognition has always been a very active and challenging research topic in the field of computer vision, but also it is one of the hot topics of the new generation of human-computer interaction technology. So the researchers have widespread attention at home and abroad. On the basis of the analysis and conclusion of domestic and foreign research, this paper research deeply on model and algorithm of vision-based hand gesture recognition. The main contents and results in the paper are the following:The detection method of hand gesture based on statistical color model of skin color is researched in this thesis. The comparative of the effects of the color clustering in the conditions of the same light and under the different lighting in RGB, HSV and YCbCr color space is analyzed. Then the ellipse model of skin color in YCb’Cr’color space is presented based on YCbCr color space. The detection method of hand gesture with this model can remove the noise of some light and is rapid and accurate.The method of extraction hand gesture outline is researched in this thesis. The method of extraction the gesture outline is proposed based on the fusion of the GVF Snake model and color model. The outline of hand is detected by this method to get the initial outline of the GVF Snake model to improve the results of the GVF Snake model. The outline is computed base on the area by detecting with skin color model to reduce computation.The detection method of the dynamic gesture is proposed based on the fusion of local background updating and skin color model. The problem of the method based Bayesian background subtraction is longer computing time and has interference area. But the foreground region of movement gestures is detected accurately and rigidly by the method of the thesis. And it is strong adaptability even in the complex background. The recognition method based on the normalized Fourier descriptors and the RBF neural network for static gesture recognition is presented.The removal method of the arm wrist area is proposed based on minimum bounding rectangle spindle position against the physical properties of hand area. The interference of the arm region is removed effectively. The minimum bounding rectangle is detected using convex hull boundary information to reduce the computation than the traditional construction method. The character of normalized Fourier descriptors is researched. Then select method of the number of low-frequency Fourier descriptors is presented by analyzing the hand gesture outline normalized energy distribution of Fourier descriptors. The recognition method of static hand gestures is proposed based on the RBF neural using normalized Fourier descriptors.Hand gesture tracking is difficult and key problem of hang gesture recognition. According to the character of Kalman filter is little calculation, real-time, unbiased and optimal features, the target tracking method of the hand gesture is proposed by using the fusion of Kalman filter and skin color model. The centroid place, minimum area enclosing rectangle width and height of hand motion region is detected by skin color model to as the basic motion parameters of moving objects of Kalman filter. The match function of the hand gesture is designed with spatial structure characteristics and color probability feature of target to ensure the accuracy of the object matching. The matching method of color probability characteristic is proposed by Bhattacharyya coefficient.The experiments show that this method could track hand gestures effectively and robustly. The human face and the part of the object of the surrounding environment which are similar to hand will cause inaccurate detection problem of hand gesture detection. The removal method is proposed based on the movement of information and labeling of the moving region.The beginning and end position segmentation method is proposed by presetting the detection area of dynamic hand gestures against the uncertain problem of gesture trajectory segmentation.Then the smooth motion curve of hand gesture is obtained by thrice spline interpolation algorithm. The moving trajectory characteristics is presented using location normalized and angle to resolve the different time and space issue of hand gesture moving trajectory. The method of calculating the gesture sequence key frame is proposed by the distance between adjacent trajectories, which are obtained through hand gesture track. And the frame of the start and the stop position are the key frame of hand gesture motion. The hand category of is recognized by calculating Fourier descriptors of the key frame. The normalized coordinates of trajectory point, this trajectory points and the angle, The hand category of the Key frame are the feature vector of hand gesture sequence. Then the method of the dynamic hand gesture recognition by HMM model is proposed. The results show that the method is effect.