Study On Several Key Technologies In Vehicle Recognition System

Intelligent Transportation System (ITS) is a system, for integrating with communication technology, control technology, sensor technology, operational research, artificial intelligent and computer technology. The aim of ITS is to make full use of the available road facilities, enhancing the security, high efficiency and comfortable of transportation system, and then improve the economical efficiency of the whole system. ITS is one popular research problem of all countries to solve the traffic jam, the traffic accident, the lack of soil and energy resource, the pollution of the environment, and the economic lose by all of them. The application of ITS will bring both economic and social benefit. The Vehicle Recognition System(VRS) have been widely applied in advanced country such as America, Japan, Europe, but in our country, the application of VRS is in research phase.Main content of this dissertation includes some novel methods of vehicle color recognition, vehicle shadow segmentation and vehicle window location. The main results can be summarized as follows: A method of fast color recognition. Vehicle color recognition is a typical pattern classification problem, the traditional statistical pattern recognition method have showed excellent recognition performance when the number of sample is more enough. It is impossible to collect more samples in real-time vehicle color recognition application because new color vehicle appears continuous. There aremore than ten color categories to be classified, so vehicle color recognition is multiple-category pattern classification in small train samples. Support vector machine (SVM) which based on statistical learning theory has been proved to be one of the preferable classifier when the a priori probability of category is unknown and the sample is small, but the classification speed of SVM is slow when the category number is more, which restrict its application in real-time vehicle color recognition. A SVM based fast vehicle color recognition algorithm is presented by reducing the number of SVM. The experimental results have shown the proposed method is superior to nearest neighbor classifier in both recognition speed and recognition accuracy with the same small sample train set. A method for improving the classification stability of SVM. Because of high classification accuracy and fast classification speed, the SVM is a better classification method in color recognition application. Once classified error, this sample will be put into train sample set, and then the classifier is trained again. The new sample will be looked as the same as the original sample of the train set in standard SVM, hi this paper, a improving fuzzy SVM method is present by different weight fuzzy punish factor of each category and different weight factor of each sample. This method can improve the classification stability and sensitivity of two-category distribution imbalance.A method for improving classification accuracy of SVM. There are shadow areas which the sample category can not be distinguished when multi-category classification is realized by combining the result of 2-category classifier such as SVM, this problem exist also in vehicle color recognition. A fuzzy logistic SVM based on fuzzy set theory is presented to solve the classification problem of shadow areas. This algorithm will not influence the decision-making boundary in non-shadow area and will improve the accuracy in shadow area.A method for improving vehicle body color recognition accuracy. When thecolor of each pixel of vehicle body is identified, we usually use voting method to decide the final color of the vehicle. The most vote number of the color is not vehicle's color because of the influence of the window and the container truck. Since the D-S evidence theory is able to solve uncertainness, a vehicle color recognition algorithm based on D-S evidence theory is presented. By reducing the influence of the window and the container truck, the accuracy of vehicle color recognition is improved. A novel method of vehicle sh...