Based On The Actual Characteristics And Models Of Fuzzy Svm Classification Method Research

In the21century, more and more country pay attention to the construction of intelligent transportation system, and the vehicle automatic classification system is one of the important component. With the development of image processing technology, pattern recognition and classification, there are more and more scheme of the vehicle recognition system, and the result of recognition are more accurate. The vehicle automatic classification system will have widely application in the urban traffic control system, emergency command, accident detection, intelligent route guidance and highway automatic charging. On the other hand, with the national economic development, there are more and more vehicle type, this factors also increases the difficulty of classification system, therefore, more and more scholars and research organizations are related to the study of the classification technology.In this paper, the mainly research object is vehicle image on highway, study on the vehicle automatic classification by using image processing and support vector machine (SVM) method. The main work is as following:(1) Vehicle image preprocessing and the improve method of vehicle feature extraction. Acquisition image from outdoor camera, and pretreatment process includes:the grey of images, the image smooth, image segmentation, the image of binary-, etc. IThrough the experiment, selected the method which have good effect. When detection the feature of vehicle outline, using scanning the pixel one by one. But some features could not find by scanning such as:the feature which can calculation the height of the car head. Therefore, in this paper using the method which combining with actual models feature value to determine a experience value, in the range of this experience value, looking for some difficult feature points by slash scanning.(2) The research of vehicle classification method. The principle of first classification is transformation pixel feature to the actual characteristic by some data such as:camera installation height and angle. Then, according to pattern matching with the sample data, the test vehicle will divided into large, medium and small three kinds of models. Using the fuzzy support vector machine (FSVM) method take the first classification result divide into the two compartments cars, three compartments cars, vans, medium-sized buses, medium-sized trucks and large passenger cars, large trucks.(3) Sample collection and classification experiment. Through the actual measurement and automobile manufacturers official web site, collect a large sample models actual data, and at the same time through feature extraction got a lot of vehicles pixel data. Then, through experiment to find appropriate kernel function and punish factor value.Finally, analyzed the experimental results.In shortly, through the analysis of experimental results, the method based on the actual feature and fuzzy support vector machine have some advantages. Such as: sample characteristic vector is not big, little data redundancy, the error rate is low, etc. So, this method can identify targeted models good, and have some application values.