Real-time vehicle classification based on eigenface

Vehicle classification is a hard task in ITS. A real-time vehicle classification method based on eigenface is proposed, it includes two main steps: training and classification. In the training step, first, using the time average image approach to obtain and update the background model, and then, using the background difference approach to detect and extract the outline of a moving vehicle, furthermore determine the left, right and bottom border of a vehicle face according to the left, right and bottom border of the vehicle outline, next, the height of a vehicle face is set to a fixed empirical height. After normalization and some other necessary preprocessing steps, the vehicle face image library is built, At last, the eigenvectors of a vehicle face image by using eigenface method are extracted and the vehicle face feature library is constructed using these eigenvectors. In the classification step, first, a vehicle face image and its eigenvector are extracted by the above ways, and then compare the difference between the vehicle face eigenvector and the eigenvectors in feature library using the minimum distance method. The experimental platform is built on OpenCV and Visual C++, the vehicle face feature library is constructed using 100 vehicles face. In the experiment, when the size of vehicle face image is 80∗30 pixels, the average time of one vehicle classification is about 1.88 ms, when the size of vehicle face image is 120∗50 and 322∗131, the time is about 3.28ms and 30.69ms respectively. The experimental results show the feasibility and real-time performance of the system, when using the train library as the test library, classification rate was 100%.