Research And Application Of Hand Gesture Recognition Technology Based On Depth Image Information

Nowadays,the development of technology can’t do without the wide application of electronic computer. In pace with the development of science and technology, we have increasingly high expectations of computer. Human-computer interaction is still one of the hot spots in research fields. With the development of ubiquitous computer technology, human-computer interaction approaches with keyboard, mouse and pen nowadays are no more sufficiency. Because of the limitation of these devices, the available instruction set as well is limited. Directly using hands as an input is a fascinating approach for providing natural human-computer interaction. Natural human-computer interaction interfaces has evolved from interfaces based on text, interfaces based on2D graph, interfaces supported multimedia, to Virtual Environment (VE) systems with multiple participants. Imagine the future for the human-computer interaction;You can install a3D application, which you can use to move objects simply by moving your hand and you don’t need to touch any input device during the whole operation processes. As long as computer vision technology develops, the image above will come true before long.In order to make hand gesture recognition that based on depth information come true, this thesis analyzes nowadays hand gesture recognition and designs a hand gesture recognition system based on depth image information. We use the approach of median filter and the border following algorithms,etc. This system is consist of two function module:Hand Posture Detection and Hand Gesture Recognition.First,the system gets the gesture informations from the Kinect and stores them. Then it segments and pre-processes those image informations and gets the hand’s digitized binary image. It model gesture with Suzuki border following algorithms and gets the hand’s contour data. There are two feature approaches to recognize the hand gesture:one based on the Convex and Concave Structure and the other one, structured matching, based on Hu Moment.This gesture recognition problem is essentially a small samples and nonlinear pattern recognition problem.In order to improve recognition rate, the system trains and tests the feature values of a plurality of gestures samples, Hu moments and convex and concave structure,by the approach of SVM. The experiment proves that the recognition rate for the "0" to "9" number gestures is more than95%.Finally, it starts human-computer interaction by calling the gesture controlling method, such as controlling PPT.