Traffic Detection Based On Visual Selective Attention Mechanism

In resent years, with the growing popularity of automobiles, the road on use cannot keep up with the increasing number of cars. Then we should study on how to use the existing transport network effectively, mitigate traffic congestion, improve road utilization and reduce traffic accidents.The application of computer vision technology to traffic detecting and information collecting is an important subject in Intelligent Transportation System, and moving vehicle detecting and recognition is the basic part. To solve the big computation load and bad timeliness of traditional digital image processing method, a visual selective attention mechanism is introduced in the moving vehicle detection. Taking the natural road as the background, a kind of moving vehicle detection and recognition method is brought forward, and moving vehicles can be automatically recognized elementarily by using this method. The main work of this thesis is described as follows:Firstly, With the study of existing vision attention models which based on the theory of vision attention, We first extract focus of attention using image-driven, bottom-up attention model proposed by Itti etc. Consider the existence of the noise and other kinds of disturbance, the extracted focus of attention is possibly an isolated point, and has not fallen in the interest region of the goal object, So we make an improvement to Itti's model, We propose a vision attention model which adjust the focus of attention according to the whole effect principle of the feature saliency map. The experiment indicates that the method can effectively shift the focus of attention.Secondly, In order to extract the vehicle from the transportion scene, based on the theory of vision attention, we propose an algorithm which integerated movement feature and visual features such as the brigtness, the color and the orientation features to detect moving vehicles in transportation scene.The experiment verifies the effectiveness of this algorithm. Thirdly, After having detected vehicles based on the visual attention mechanism, In order to count on the transportation traffic flow, we set a detecting line on each traffic lane, and count the vehichles which through the line. Moreover, In order to recognize the vehichles, Based on analysis of many object recognition algorithms, we present an algorithm which fusing mang features to recognize objects, it integrates the vehicle's edge line length, the complexity of the sharp and other features to recognize objects. The result shows that this method can distinguish large-scale, medium and the small vehicles effectively, and the rate of accuracy is quite high.