The Research Of Static Gesture Recognition Based On Hu Invariant Moments And Stucture Features

The possibility that the gesture will be used as a computer input method has been increased in the futuer because of its intuition and consistent with the behavior characteristics based on the research of human-computer interaction(HCI). Using the gesture as the input of HCI can comletely subvert the traditional one which must depend on the external devices.However, there can be different interpretations of the same gestures in different application backgrounds and also the hand itself is an extremely soft and deformable body, all these problems will increase the difficulty of extracting features of gestures.This thesis discusses the development of the existing gesture recognition technology and the relevant technologies. And it focuses on studing the key techniques of gesture recognition, such as fingertip detection and feature extraction, using a gesture recognition method based on the combination of anti-distortion feature and support vector machine, the effectiveness of the proposed algorithm is verified by the recognition effect of five predefined gestures.Firstly, in this thesis, we collect image frames of video stream from Kinect are used as the samples, and gesture sample database are established by preprocessing. Secondly, in the stage of gesture segmentation, the segmentation threshold is set up by using depth information and palm position, and the adaptive and robust of gesture segmentation is improved. Third, in the feature extraction stage, the distortion feature is selected, and the rotation and scaling of the gesture are extracted with the feature extraction. Finally, in the step of gesture recognition, according to the actual situation of the sample collected in this thesis, we choose the support vector machine which based on radial basis function as the classifier to test.The main contribution of this thesis is in these three parts: gesture segmentation, fingertip detection and feature extraction:1、This thesis presents a fingertip detection algorithm. According to the intersection of the convex hull and the rough classification of hand contour point fingers, combination the vector angle of composition with the sequence of the convex defect points, then determine the correct fingertip point. Furthermore, the structure of the algorithm is described in this thesis.2、This thesis proposes a feature extraction method of static gesture. After experimental comparison of the effectiveness of the feature combinations, the final choice is based on the 7 Hu moment invariants to add the number of fingers, to constitute a set of anti deformation feature vector as the feature of the static gesture recognition in this thesis. By comparing with several other methods, the feature extraction method can be used to recognize the deformation gesture and guarantee the average recognition rate of the deformed gestures.