Hands Gesture Recognition Technology Research Based On Monocular Vision

Gesture recognition research has a very important theoretical significance and a very broad application prospect. The past hand gesture recognition research, however, was mainly one-handed. One hand’s dimension of the feature vector is less than double hands’, so that it conveys less information generally. Existing double-hands gesture recognition research is strict to hardware devices. Therefore, unarmed, easy, low-cost double-hands gesture recognition research has important practical significance. This article researches on the main process of monocular double-hands gesture recognition, including gesture image pretreatment, feature parameter extraction, and hand gesture recognition. Besides, achieves the recognition of five one-hand gestures and fifteen double-hands which were input from ordinary camera under Microsoft Visual Studio2010, and builds a simple gesture interactive system, applying to one-hand and double-hands interactive systems.Collect gesture images from video stream obtained from common camera, and pretreat gesture images. In regard to interference on binary image after pretreatment, it uses CamShift algorithm to track and locate gestures. Taking advantage of the search box to search exact location of double hands in gesture images, it limits the processing areas of images in the search box, reduce the processing area of images, and improve the identified efficiency of the system. For the gesture tracing lost or the tracking box can’t contain the exact gesture areas completely, it use "search window" to cut the gesture tracking box to optimize the processing. Then, the tracking box can contain the exact gesture areas completely, reduce the number of processing pixel, and improve the efficiency and accuracy of identification. After determine the exact position of the double hands in the image, it extracts the pixel distance distribution feature with translation, rotation, scaling invariance of the gesture binary images. It can not only describe the hand gestures’ overall shape information, but also exclude the interference of the rotation and scaling of hands in the gesture feature extraction process effectively. Finally, it uses dynamic template matching identification method to identify the gestures, makes the obtained gesture features compare with the gesture features in the model library, and selects the gestures that are familiar with users’ gesture feature mostly. Finally, it uses gestures to interact with computers simply. When computer has recognized gestures, it can display corresponding gesture images on the display and feedback identification information to users. This article mainly focuses on two aspects, including the realization of hand gesture recognition based on monocular and the application of the human-computer interaction. The originalities of this article mainly display in the following two aspects: firstly, before extraction of gesture feature, it uses the method of gesture tracking and positioning to select gesture areas, determines the exact position of the double hands in the image, reduce the number of processing pixel, and improve the efficiency and accuracy of identification. Secondly, compared with the past hand gesture recognition research, the system achieves double-hands gesture interaction recognition with one camera.