Researchand Application Of Gesture Recognition Technology On Kinect

With the development of computer technology, HCI has undergone a series of development: Command Line Interface, Graphical User Interface and a natural Human Machine Interface(NUI). In the Natural User Interface, Users communicate with a PC by a natural way, such as voice, touch, gesture, etc. Gestures are a natural and intuitive interaction model. Vision-based gesture recognition is the next generation of human-computer interaction as an integral part of the key technologies. Kinect,as a revolutionary product,provides new ways to interact and more deeply reveals the concept of interactive human-computer interaction. It is able to capture, track and analyze body movements, gestures, sounds and more types of body posture. However, Kinect gesture recognition accuracy could be improved.To solve the problems, first of all, the article presents a Kinect-based hand tracking the trajectory optimization, using a Kalman filter for image stabilization. Kalman filter is an efficient autoregressive filter, it can be from a series of incomplete and contain noise measurements, the accurate estimation of the state of the dynamic system; under Kalman filter can also measure the variance is known, from a series of data in the presence of noise, the predicted state of the dynamic system. Secondly, the proposed based on the RGB Skeleton Hand accurate segmentation, in order to get a more accurate hand segmentation map, this paper combines the depth and Skeleton information presented combines the depth of hand segmentation. Thirdly, SVM to solve the classification problem under the small sample, the final decision function with only a small number of support vectors related to a small number of key concern of samples also improves the robustness of the algorithm.Based on the study of these three key issues, this paper designed and implemented a system based on human-computer interaction Kinect drawing three-dimensional space, to achieve a draw conducted in three-dimensional space, real-time identification palm fist, release, rotation, hands open or close together, etc. gesture, for initial control of the respective model, to terminate, view, zoom in or zoom out operation. System integration we propose a new approach, with a simple user interface, bringing a whole new user experience.