Research On Vehicle Model Recognition Technology In Intelligent Transportation System

With the rise of car penetration rate, the traffic problem became more serious, and the traffic management became more difficult, the modern intelligent transportation system is made. The Vehicle model recognition technology is one of the most important technology in intelligent transportation system, mainly through the vehicle model recognition system to collect vehicles original image, using the image processing algorithm to analysis, process and obtain vehicles information, finally doing some intelligent data management.This paper mainly researches the vehicle model recognition system which based on streaming video. At the testing of movement vehicles, it proposes the background image method based on the color image difference and background image updating method to ensure the moving vehicle area; and using a median filter algorithm and vehicles of gray histogram based on adaptive threshold value method to obtain the ideal model of binary segmentation image; and then choose the morphology processing method to regional fill, eliminate and boundary extract vehicles image area, obtain the vehicle model outline images.In the stage of vehicle feature extraction, this paper chose to use moment parameters as models feature for the vehicle recognition. Choosing to extract the moment invariant characteristic features and the improved pseudo Zernike moment characteristic features. Through the contrast experiments found that comparing with the invariant moment, improved pseudo Zernike moment characteristic features has more advantages and practicability.In the stage of vehicle model recognition, this paper introduce the pattern recognition and the related statistics theory, and based on analysissing the support vector machine theory and its classification, combined with the characteristics of sample models library, on the MATLAB simulation platform design a SVM classifier system based on the polynomial of kernel function, through experiment ascertains the nuclear function parameters of the classifier, and in the final vehicle recognition has a higher recognition rate. At the same time, through designing some contrast experiments to verify selection method in this paper which has superiority. First of all, the experiment which between the invariant moment and the improved pseudo Zernike moment characteristic features, and verifying the improved pseudo Zernike moment has a better advantages; and then through testing seven moment invariant characteristic features and seven improved pseudo Zernike moment characteristic features of different combination, get a group of two kinds of characteristic combination with a high recognition rate; by using the BP neural network and the SVM classifier to classify recognition according to the same vehicle model characteristic, finally, we find that both in recognition rate or in training classifier needed time, SVM classifier has advantage, and gets a good results.In the last part of this paper, which introduces the hardware and software environment of the vehicle recognition system, at the same time, through using the C++language and MFC framework design method to design a basic way video image acquisition system, which will be able to realize the simple function of video image collection.