Feature-based Vehicle Classification Study

As the development of information technology, communications technology, pattern recognition and image processing technology and so on, Intelligent Transportation Systems (ITS) develop rapidly. Most of the exiting traffic classifications based on video are achieved from the side. In the practice, the camera is easily obscured by pedestrian, and the vehicles overlap and occlusion have bad effect on the accurate classification. Therefore, this paper proposes a new method for classifying vehicles, which based on the vehicle features and support vector machine.In the method of the vehicles classification based on features and support vector machine, video capture devices are installed above on the road and along the direction of the road. This method not only avoids the problems of vehicles overlap and occlusion, but also could classify vehicles by these visual features. At the same time, classification based on features could save much time. Compared to estimating the length and width of the vehicle, it is much faster. In the program, if the vehicle does not have visual features, the parameters of the vehicle would be estimated and then we classify it by the support vector machine.Features extraction is very important in the vehicle classification. In the experiment, the detected vehicle should be extracted completely. In order to reduce the impact of noise and exclude the impact of color on the classification, the extracted image has to be filtered and grayed. The gray image also could save storage space. Then we detect the comer points on the grayscale. It easy to find that the area of the double wing mirrors gathers a lot of corners. The middle class vehicle could get this feature easily. But the big class vehicle's length is very long, the double wing mirrors are hard to find. And the small class vehicle is verv different from the middle class. Some of the middle class vehicles have aerial on the top of its body. After detecting corners, the aerial would appear as a single point. Other class vehicles do not have this feature. Therefore, those features could be thought as the symbol of the middle class. In the program of classification, we find the features of the vehicle firstly. If it has, this vehicle will be classified to middle class. Otherwise, the simulation curve would be draw around the vehicle according the corner points, then adjusting the curve properly to estimating the length and width of the vehicle in order to classify it to corresponding class by support vector machine which has been trained well. The experimental results have proved this method has high classification accuracy, and the classification based features speed up the classifying process.