Research On The Gesture Recognition System Based On Stereo-vision

With the rapid development of information technology, interaction between human and computer is increasingly becoming an indispensable part of the daily life. The technology in man-machine interface is more mature than before. Although traditional method of interacting between human and computer such as mouse and keyboard is directing to the more convenience, the freedom and naturalness is still not enough. A new technology which can meet the habits of the daily life's interactive communication is more and more popular. So a vision-based method for interaction has gradually become a research focus in recent years, this method can make the interaction more natural and more human-like. This technology will have a wide range of applications in appliance control, robot control, sign language teaching, game control and so on.Just because the hand itself is not rigid, so the hand gestures has the characters of multi- mode and multi-meaning, even the same gestures made by the same person in a different time are not exactly the same. Gesture recognition involves a wide range of subjects including digital image processing, pattern recognition, artificial intelligence, computer vision and other subjects, so vision-based gesture recognition is a challenging research topic.Gesture is divided into two kinds:static gestures and dynamic gestures. The static gesture means the hand's shape at a time point. The dynamic gesture refers to the trajectory generated by the hand when moves in space.This paper mainly studies the segmentation and recognition of the static and dynamic hand gesture based on computer vision system. Just because the two-dimensional information collected by camera is projected by the three-dimensional space objects, sometimes losing some information is inevitable. Image information collected by a single camera may not be comprehensive, in order to make the image information can reflect space objects more similarly and realistically, two cameras was used in this system to collect the image information from different perspectives to reduce the incidence of false consciousness.The research paper includes the following:Firstly, in the gesture segmentation, compare some commonly used segmentation method. Through back ground difference, we can find out the motion region, and with the skin color information, we can divide the hand gesture from the image. At last, some post-processing just like filling cavities, removing isolated noise points is needed to make the gesture more accurately segmented.Secondly, in the part of feature extraction, common gesture features and feature extraction methods were studied in detail in this paper, including geometric invariant moments, regional focus, the number of fingers and other features. As the gestures discussed in this issue is relatively simple, we choose the numbers of finger as the feature of static hand, and choose the center point of the hand gesture's region and numbers of fingers the number as the dynamic gesture's feature vectors.Thirdly, in the terms of gesture recognition, the recognition process and methods were discussed in this part, and Hidden Markov Model was used to implement the training and gesture recognition.