Vehicle-logo Recognition Based On PCA And SVM

With the rapid progress of the economic and the development of industrial technology, people’s living standards gradually improved, the huge traffic pressure resulted from various vehicles which are increasing year by year, problems of highway traffic appeared increasingly. We can solve these problems effectively by the intelligent transportation applications.As an essential part of intelligent transportation system, vehicle-logo recognition system has a wide range of applications in no parking charge system, automatic recording of crossing, and parking lot which nobody safeguards, criminal investigation, etc.Vehicle-logo recognition technology as the core of the vehicle recognition system, has become one of the most important point on international. More and more scientists and researchers are focus on this technology. This technology takes digital image or video signal flow which are collected by the camera, as the object of study, to deal with these objects by the image processing and automatic recognition technology. Vehicle-logo recognition technology mainly includes two key technologies of vehicle-logo positioning and vehicle-logo recognition. This paper aims to the study of vehicle-logo recognition method.Through the study of the method of vehicle-logo recognition, in order to solve the problem that existing vehicle-logo recognition methods more depend on acquisition environment and personal experience, this paper chooses support vector machine classifier instead of the neural network classifier which is now commonly used. For support vector machine classifier avoids "learn" phenomenon which may occurs in the training process and improves generalization ability. The experimental results show that, the logo data in C-SVM model can achieve good classification results when after PCA algorithm dimension reduction.This paper puts forward a new vehicle-logo recognition method combined with the least square method which can further enhance the recognition rate algorithm. This method firstly uses the smoothing filtering technique to remove noise, and then apply PCA technology data dimension reduction to obtain feature vectors which can reconstruct the logo image, finally using the least square support vector machine (LS-SVM) to identify logo. It is proved that the method can obtain satisfactory recognition rate, higher operation rate, and strong robustness by samples experiments of different time and light vehicle-logo. Compared with SVM classifier, the method is more suitable for real-time vehicle-logo recognition.This paper designs a vehicle-logo recognition, and a vehicle-logo query system, showing the systems processes. Then simulate them by MATLAB 7.11.0platform.