Gesture Recognition Based On Dynamic Bayesian Network Model

With the development of computer technology, more and more techniques are applied to achieve human-computer interaction. People hope the input information is simple and direct,also,the system to control the computer is consummate. In life, communication among people mainly are by the way of natural language and body language. Because the body language is affected very little by different race and region, more and more researchers focus on using human body language as the input information and building the appropriate system to achieve human-computer interaction. Gesture is a simple body language, and it can express basic human intention, so the gesture as an object of study can be used to achieve human-computer interaction. The description of gesture is affected by hand shape, posture and position. The gesture recognition systems have different broad applicability because of the different characteristic information and different recognition algorithm.In this thesis, the work is that the gesture recognition based on dynamic Bayesian network model. The specific work will be described as follows:(1) According to discussing for development and current situation of gesture recognition, its future is identified. The structure and works of this thesis is expressed in detail here.(2) Studying the methods of gesture positioning and comparing the advantages and disadvantages among each method,it choose a skin color template to achieve the location by building a better color template. The skin area can be located by morphology processing and threshold segmentation.(3) Calculating the centers of skin areas, they can represent hands and face.It track the centers based on the optical flow.Then,it calculate the gesture motion vectors and obtain the gesture chain codes by vector quantization and coding.(4) According to analyzing the characteristic chain code, a suitable Dynamic Bayesian Network model whose input is code chains is built, it is fulfilled by inference and learning.(5) According to analyzing to a plenty of experimental results, the validity and feasibility of the proposed entire structure system of gesture recognition algorithm based on Dynamic Bayesian Network model is verified in this paper.According to a large amount of experimental results and statistics, the method mentioned in the thesis is more universal and more accurate.