Research On Vision-based Static Hand Gestures Recognition

With the rapid development of virtual reality technology, research on new human-computer interaction technology is extremely active. The research includes face recognition, expression recognition, hand gesture recognition, pose recognition and so on. Among them, gesture recognition is tipped to a great demand.The writer studied the key algorithm proposed in this paper. The technique essentially includes video detection and segmentation, image preprocessing, feature extraction, and recognition of a number of gestures which are assigned as some control commands. The specific tasks are as follows:(1) To get the aspects of gesture images, the writer used the VFW (Video for Windows) API function. Specifically, the writer used the news AVICAP module, the macro function, structure and the callback function which is used to capture the real-time video to receive the video custom frame stored to the data structure in DIBINFO.(2) In gesture image preprocessing, the writer first used histogram image to binarize the gesture image, then according to the algorithm proposed by this paper, a composite image was denoised, then the writer hollowed out the image using the method of the internal point gesture target contour extraction, and finally the writer used the eight-neighbor method to scan the border gesture tracking targets.(3) In the process of signing aspects of image feature extraction, the writer studied the characteristics of commonly used gestures, such as staff size, staff profile of the rectangular area, the fingers '(except thumb) number and so on, and then in accordance with the characteristics of neural network, the writer proposed the images' grid feature extraction methods.(4) Finally, the writer used the hybrid method to recognize the gestures. First, the writer signed the common characteristics of rough image classification. Second, the writer used BP neural network algorithm to identify the images. In this paper, 10 kinds of common hand gestures were put as standard gestures in the image database when training samples. In the recognizing process, this writer put the sign of the images' grid feature in the input layer. In the hidden layer the writer used sigmoid function, and in the output layer the writer used 4 nodes.The results showed that the static gesture image recognition system has an effect on binarizing images and denoising effects. With the relatively simple background, the system recognition accuracy is better. However, with the context of more complex cases, the effect of system identification is not obvious. Therefore, the gesture image recognition technology proposed by this paper is only to improve the existing methods and the merits of the algorithm remains to improve.