Large Screen Human—computer Interaction Relate Technology Research

While large Screen multimedia display system is increasingly being used in various occasions, such as science and technology museum, exhibition hall and other places, there isn’t a perfect solution for large screen human-computer interaction. Gesture-based interaction has become a research focus in large screen human-computer interaction program, as a result of the advantages of natural operation, well experience and more in line with the requirements of human interaction.Large screen human-computer interaction mainly involves three aspects of content, man-machine interactive human body tracking, gesture interaction tracking and identification, as well as the interaction involving text input problem. The main contributions of this paper are listed as follows:Firstly, this paper proposed a fast mechanism of updating background based on Gaussian background model for object detection. In the case of instantaneous illumination changing, Gaussian background lead to the problem of foreground error detection because of slow speed of updating background. The method in this paper has resolved it by using the method of combining the background subtraction and gaussian mixture model.Secondly, this paper has discussed and improved the tracking algorithm based on Meanshift,and improved its robustness. Aiming at the bad robustness,leaded by building model with the single color feature and single whole target area of Meanshift, in the situation of target scale changing and light changing, this paper presents a model based on modules and the features of texture and color in each model, so that improves the robustness of the algorithm greatly. It also discusses the algorithm of body tracking under the condition of body block in this paper. Combining with Kalman filter prediction algorithm, a good tracking effect is obtained in the case of body blocked completely.Thirdly, this paper presents a simple method of binocular vision positioning. In traditional binocular vision positioning method, the way of camera demarcation is complex and is unfit for actual situations. But in precision, we can use the camera projection geometry and spatial linear relationship to improve the computational efficiency, and make it convenient for practical use.Forthly, this paper puts forward a way of static gesture recognition to input Chinese characters to the system according to Pinyin input method. When we collect gestures with traditional RGB camera, they are often effected by inertia, complex background and lighting. We can avoid it by using depth image to segment image and get a static gesture, so that it improves the robustness. Extracting the static gesture feature with the way of extracting features of sift, on one hand, it retains the contours of all kinds of ratating and keeps zooming deformation, and stregthens local regional characteristics and improves the recognition rate on the other hand. The method of gesture classification is based on SVM. Constructing a classification tree combining with the rules of Pinyin, we can provide a way to input ten Chinese characters per minute and can effectively meet the needs of searching, where we want to input key Chinese characters.Finally, in the dynamic gesture interaction, dynamic gesture tracking with particle filter can deal with nonlinear target and characteristics of non-Gauss distribution system, using particle filter algorithm to track the gesture for gesture trajectory. The color feature is generally used for the traditional particle filter, and the partic’ number is fixed. This paper uses gesture contour line according to the current state of motion gestures to dynamically determine the number of particles, and prevent the particles from degradation by the dynamic allocation of the particle weights. Get the gesture trajectories, the discrete quantization using16direction angle as input vector quantization of the gesture recognition. Since the Hidden Markov models (HMM) can be related to space and time model, it can be suitable for dynamic gesture recognition. Although the number of traditional HMM initial states is based on the value of experience, this paper uses the key point algorithm for gesture with different initial states to set different values, which makes the initial state setting have a reference mechanism. In the selection of training samples, using chaos algorithm to find the best training samples can make the HMM model is trained have the global optimal characteristics. Using the method of threshold to improve the traditional HMM method make the HMM have certain rejection rate, exclude the error gesture input, improve the recognition rate, prove that the method is effective, and be interactive.As verified by experiments, this paper solves the three problems appeared in large screen multimedia interaction effectively. So the users can get better and more natural experiences during the process of interaction with large screen.