Research The Vehicle Detection And Classification In Intelligent Transportation System

Intelligent Transport Systems (ITS) is the determinate approaches for the development of the modern transports, the vehicle detection and classification based on computer vision that is a important research field for the advance of the ITS。 It has promising prospect in the application of the road traffic surveillance system and the highway toll system and so on. In this thesis, we research deeply the technique about vehicle detection and classification, and propose a precise and robust algorithm of detecting and classifying moving vehicles on road. The main contents can be listed as follows:First, this thesis presents a background reconstruction algorithm based on median pixel intensity classification after analyzing and concluding many kinds of vehicle detection algorithms. It can detect moving vehicles precisely by background subtraction. And then, considering the relationship between the pixels, the algorithm divided the background into several blocks by means of the statistic table of pixels change, and used the different velocity updating every block, which can make the model adapt the changing environment quickly. To solve the shadow problem, after the underlying physics of shadows and their characteristics are studied, one shadow segmented approach is exploited. It transforms the subtractive image from RGB color space to SRG color space. The experiment shows that it is efficient to segment the shadow with the moving vehicles.Secondly, in the vehicle classification module, we process the subtractive image by the morphological filtering and connected component labeling, obtain the figure and position of moving vehicle, and decide the characteristic vectors.At last, two vehicle classifiers are designed, one is the RBFNN classifier based on fuzzy K-means clustering, the other is the RBFNN classifier based on immune algorithm. Under the same experimental condition, the classified result shows that the latter is better than the former in reliability and speed.The methods proposed in this paper adapt to complex scenes such as large area and multiple objects, and can satisfy the requirement of vehicle detection and classification in natural environment. The research would be reasonable and valuable in theoretical and practical areas, and can be generalized to other fields of video surveillance.