Research Of Opencv-based Natural Gesture Recognition And Interactive System

With the development of virtual reality technology, the freedom of operation, intelligence interactive become the next generation human computer interaction technology trends. It is people-oriented, the computer as a multi-channel, multi-mode, multi-media’s perception of identification device, identify the people’s voice and action, including human faces, gestures, body potential expression to achieve interaction, than the traditional mouse keyboard interactive way to a more natural convenient, more free than the touch-screen technology in today’s flexible, hardware cost more low, interactive efficiency higher. Therefore, demand for such a natural and harmonious interaction, the voice, face, gesture recognition is an important issue, identify the effect of experience is critical for interactive applications. This article in view of the lower adaptability and accuracy of original world expo guangzhou museum "360degrees virtual interactive experience area" project action induction solutions, in-depth study of gesture recognition and interaction, not joined to the infrared light, using only the camera case, extend the function and application of the original motion-sensing program.This paper in-depth analyzes the gesture recognition technology at home and abroad and the research situation, summarized the basic concept, the basic theory and the main technology and key problems, Based on this, the thorough research and design natural gesture recognition method, mainly has five part:First, image preprocessing, including image filtering and image noise reduction; The second is the color model, including the choice of color space and color model initialization; The third is the gesture motion tracking, on the basis of the Mean Shift Algorithm and CamShift algorithm to improve the Camshift filtering algorithm; Fourth, the gesture recognition, using two-dimensional template matching contour; Fifth, the gesture feature extraction, including the calculation of the gesture area, perimeter, direction and fingers to determine; Which focuses on the design and implementation of algorithms, including:adaptive skin color modeling and segmentation algorithm, adaptive mean shift tracking algorithm, Hu moments contour matching algorithm, Arm removed and fingertip extraction algorithm.In order to verify the identification algorithm proposed in this paper, This paper is designed a natural gesture recognition and interactive systems based on the OpenCV library. The system can recognize the ten predefined static gestures, and users through the template definitions function custom gestures, and extracting the gestures feature points, including the calculation of gestures area, perimeter, movement direction and fingers judgement. System design multiple interactive applications, including fingers drawing and through the gestures view picture and control the mouse, and realize the dynamic application interaction between user gestures and computer. Through the system test shows that, the recognition rate of system identification and speed achieve the expected.Characteristics of this thesis are reflected in two aspects:(1) To improve the adaptability of skin color model, this paper puts forward a compatible YCbCr and HSV space color of skin modeling method. This method can get the system or users of the real-time capture predefined color samples, will the gaussian model of the generation of skin probability value set to the initial value in every color space, and through an adaptive mechanism to find the best color threshold. Finally, combined with a variety of image processing algorithm to realize color segmentation. The experiment shows that this method increases the environment on the change of the correctness of the recognition gestures.(2) Existing in the environmental interference tracking method cannot be accurately track the target, put forward a kind of combination Kalman filtering improved Camshift algorithm. First, the current frame image Kalman predict the location of the object, through the Camshift algorithm to search and target the most similar goals and template updating Kalman filtering state. The method makes search window can target outline with change and real-time update, improve tracking the adaptability and robustness.