The Recognition And Application Of Static Gesture Based On Hu Momtents And SVM

With the continuous development of Image processing, Pattern Recognition, intelligent computer and some other related technology, gesture recognition gradually become the focus of people’s discussion. Due to the gesture’s characteristic of diversity, ambiguity and the otherness between time and space, and so on, gesture recognition proves to be a multi-discipline research topic full of challenge. As a spacial way of recognizing creature’s characteristic, compared to fingerprint and face recognition, gesture recognition is more friendly and convenient, and has more applications in man-machine interaction.This text summarizes and introduces the existing technology of gesture recognition, also its evolution, the key thing of gesture recognition’s research. After going into the image preprocessing of gesture, the extraction of characteristic of gesture, the tracking of gesture and the recognition of gesture, it comes to the conclusion of puting both characteristic based on HU moments and SVM into gesture recognition. With the help of OpenCV, it completes the recognition of predefined ten types gesture from0to9, on the platform of Code::Blocks, using C language to program a system of static gesture recognition under simple background.First of all, we got files of video streaming from camera as the sample, and built sample gesture library after preprocessing. Then, at the stage of extracting gesture’s characteristic,we mainly used Hu moments as the characteristic of gestures to recognize, and took advantage of geometric moment’s feature of stay invariant while rotating, translating the image, and changing the size of image, to solve the uncertain of scale while rotating. Next, during the track of gesture, we chose and realized the Camshift algorithm, in order to improve real-time and robustness of gesture tracking, at the same time realized the exact real-time track of gesture when read files of video streaming from camera. At last, we especially studied the support vector machine classifier when recognize the gesture. The support vector machine is a method of Pattern Recognition based on structural risk minimization, and has many special advantages on solving small sample, nonlinear and high dimension pattern recognition. Finally, we selected support vector machine tree based on RBF kernel function, and got the optimum parameters through calculating.At the end of this text, we illustrated the whole program design about gesture recognition system and the confirmation, detection of algorithm. The simulation result shows that this system can achieve desired recognition effect, and the correct recognition rate generally reach more than93%, in learning mode, video mode and camera mode to recognize ten gestures. In sufficient sunlight cases, both the robustness and stability of system obtained desired results, and the algorithm of this gesture recognition proved to be steady and effective.The innovation points of this text are as follows:on the one hand, we adopt man-machine interaction to track and aided set the gesture captured from camera, and project the gesture on a virtual plane and show it on the computer real-timely, to improve the efficiency of gesture recognition; On the other hand, we study the algorithm of gesture recognition based on SVM, and choose support vector machine tree based on RBF kernel function to enhance the precision and efficiency of gesture recognition.