Research On Traffic State Recognition Based On Vision

With rapid economic development and accelerated urbanization, urban population and amount of vehicle are continuous increasing, which lead to the more and more severe traffic congestion. Furthermore, the traffic congestion often causes problems of environmental pollution, traffic accidents and energy waste; eventually these problems badly restrict our urban development and affect our living quality. Nowadays in the range of worldwide, the traffic congestion has become a critical problem. Therefore the use of Intelligent Transport System to solve the traffic congestion has a great practical significance.On the premise that the amount of Traffic participants cannot be unlimited restricted and the capacity of the road network cannot be unlimited increased, improving the operational efficiency of the road network becomes the key method to relieve traffic congestion. It has been proven that Traffic guidance system can balance the load of road network, and it is one of the effective methods to improve the operational efficiency of the road network. Based on the current traffic flow of road network, Traffic Guidance System, provides a driving route guidance for traffic participants, adjusts the load of traffic-intensive areas of road network, reduces the probability of traffic congestion, ultimately relieves traffic congestion. Real-time traffic flow state information is the infrastructure data of traffic guidance system. And only in the availability of real-time traffic status information, traffic guidance information system can fully play its role and ease traffic congestion.With gradual cost reductions of video information acquisition, video information transmission, video information storage and video information analysis; the large-scale application of video analysis has become possible. Therefore, to achieve traffic state’s automatic recognition by analyzing traffic video, can provide important supporting data, for traffic guidance and information system-------Real-Time Traffic Flow State Information, which has great practical significance to relieve traffic congestion. Based on the Conversion and Unified Theory of Artificial Intelligence, the paper makes a deep study and research related to automatic recognition of visual traffic state, trying to establish a feasible technology and proposal. The main works are as follows:1) Base on the Comprehensive Information theory and the Information Conversion theory, to conduct a comprehensive research and analysis on the traffic parameters. From the angles of Syntactic Information, Pragmatic Information, and Pragmatic Information; to make a fully informational and significant analysis to traffic parameters, to assess the Possibility and the Utility of different traffic parameters, and to choose certain occupation ratio and instantaneous with high Possibility and Utility, as the parameters for describe the traffic state. Since it is complex to accurately extract occupation ratio, after a study of occupation ratio, the paper proposes the mean gray value of background subtracted image in observation area is as one characteristic of the number of vehicles on road, and define this characteristic as gray-scale parameters. Theoretical analysis and experimental results show that statistically there is a linear correlation between Gray Parameter and Occupation Ratio. According to the definition of gray parameters, gray parameter is unrelated to vehicle’s area, thus it could avoid complex algorithms in the parameter extraction process, to improve computational efficiency. In the purpose of overcoming rejections of different camera parameters, the paper proposes the ratio of moving vehicle’Pixel Distance to lane’s pixel length, to be one speed parameter.2) To extract the gray parameters needs to build a road background image, on the basis of Group-Based the Histogram Background Model, the paper designs a simple structured, slightly inaccurate systematical Gaussian group-Based Adaptive the Background model, which can also adaptively regulate background updating speed. With advantages of fewer adjustable parameters, faster tracking speed, and high accuracy, this model can from traffic video efficiently estimate background image and follow any changes of ambient light. Based on method of image Texture match, the paper designs the detection method of corners on vehicle, to avoid any segments to vehicles in the process of vehicle corners extraction; eventually achieves a traffic parameter extraction without any segments to vehicles.3) Based on the knowledge Theory of Conversion and Information-Knowledge-Intelligence:The Conversions and Unified Theory, the paper analyzes the method of knowledge acquisition, and utilizing RBF neural network and fuzzy set designs a traffic state classifier, which achieves high accurate recognition results. The paper defines the specific lighting conditions, traffic site traffic scene. Different lighting conditions in daytime, evening and night, result in different typical ambient brightness, thus in different traffic ambient using same classifier to classify traffic state would achieve a low accuracy. To overcome the above problems, the paper studies traffic state classification, designs all-day traffic state recognition method, and achieves an all-day traffic state recognition with a higher accuracy.Based on the knowledge Theory of Conversion and Information-Knowledge-Intelligence Conversions and Unified Theory, the paper analyzes the method of knowledge acquisition, and utilizing RBF neural network and fuzzy set designs a traffic state classifier, which achieves high accurate recognition results. The paper defines the specific lighting conditions, traffic site traffic scene. Different lighting conditions in daytime, evening and night, result in different typical ambient brightness, thus in different traffic ambient using same classifier to classify traffic state would achieve a low accuracy. To overcome the above problems, the paper studies traffic state classification, designs all-day traffic state recognition method, and achieves an all-day traffic state recognition with a higher accuracy.