Research On Automatic Classification Method Of Vehicle Models Based On SVM

With the rapid development of economy, the amount of vehicles increases enormously, problem of traffic jam becomes increasingly prominent. Therefore, how to relieve the traffic pressure is urgent need to resolve by local traffic departments.Intelligent transportation system has been more and more widely used with the progress of information technology. The transportation based on video, image processing technique and internet is playing a key role in urban development, urban management, traffic planning and guidance. However, the intelligent technique level is still very low, even though cameras cover every corner of cities and provide a large number of images, the way to obtain information is still through artificially watching videos, which leads to lose much information. In order to get as much information from the video as possible and to play the maximum effectiveness of intelligent traffic,the only way is to filter out useful pictures automatically through dynamic video image processing. The vehicle type automatic recognition technology is studied for the sake of intelligently classifying vehicles from the dynamic vehicles flow.Vehicle automatic recognition based on pattern recognition technique classifies vehicles automatically through the extraction of image feature vector method. Potential applications are obvious in many points such as none-parking charge, searching traffic information quickly, and counting up related data. However, the challenges of vehicle automatic recognition technology are the problem of accuracy and consuming time.The study aims to improve classification accuracy and to cut down processing time.The following research is conducted in this paper: a car face image template library is built with vehicle images extract each template orientation gradient histogram feature, treat vehicle images, then locate the vehicle's position, after which the histogram direction images of the bus face are extracted. The classification resultsare to be obtained by the match of each image and eigenvalues stored in template library. For the first time, the study proposed the classification method, in which feature value matching in the classification improves the accuracy of the classification,but the classification time is increased. The principal component analysis is also used for the first time, which shortens the consumption of classification time. The method which is similar to the process of multi-process, is designed to improve the operation efficiency of the program and to break the limitation of the slow computer single process operation. In this paper, the recognition rate of the feature value matching method can reach more than 90%, which is better than the current classification method, and the classification time of each image using is about 50 ms. The feasibility of the automatic identification system is proved.