The Realization Of Human-Computer Janken System

Man-computer janken system is a simple man-computer interaction system. It changes one of the janken game of people into computer, so people can play janken games free without other people. These years, as the development of technique of Man-computer interaction, the survey of ways of Man-computer interaction which matches the interpersonal communication has improved quickly. The main way of survey including:face recognition, facial expression recognition, head moving tracking, hand gesture recognition, and so on. Hand gesture recognition is a very important way of man-computer interaction because it is direct viewing, simple and easily realized. Now in many games people try to use hand gesture instead of keyboard and the mouse, and so people can be more easy to immerse in the game. But hand gesture recognition is a topic which is challenging and involving several subjects, because of the diversity, ambiguity, flexibility of hand gesture and the complexity of computer vision.This text designs a man-computer interaction system based on moving hand gesture recognition. It includes:video collection, frame separation, get useful frame, image segmentation, feature extraction, image recognition, man-computer interaction. First, we get our video by ForeSite-EDU camera, and then we get continuous image frames by VFW soft tool package. The following step, we judge if the hand gesture occurs by background deduction, and if we detect the hand gestures, we select Temporal Difference between consecutive frames. Now if we get binary images frames of which all pixels are black from continuous several image frames, we choose a image frame free as our useful image frame. To the useful image frame we get, then we separate the hand area based on H value of HSV model according to clustering character of H value of human skin color, and get the binary image. At last we get the contour by exploring method. At the characters collection phase, the character vector is made up of the eight geometrical characters from the binary image and the contour image. When we begin to recognize the image, we select pattern matching based on the minimal Euclidean distance. At last we compare the recognition result and the result that the computer gives random (1 stands for scissors,2 stands for cloth wrapper,3stands for hammer), and display the comparing results. This article takes 15 pieces of video, and each kind of hand gesture includes 5 pieces. Every piece of video is made up of 100 image frames. The average recognition ratio can reach 98.7%, which can meet our demand for recognition. Besides, the speed can also be ok in the interactive phase.