Research On Recognition Algorithm Of Vehicle Logo Under Complex Environment Based On KPCA And GPC

With the rapid growth of Chinese economy and the increasing improvement of people’s living standard, cars have become the most commonly used transportation in people’s daily life. However, the sharply rising number of cars brought new challenges to the safety management of urban traffic; vehicle identification technology has become an important topic in the field of intelligent transportation research. Vehicle logo recognition is one of the core contents for vehicle identification. In consideration of some acts of malfeasance, such as covering license plates, paraphrasing license plates and so on, recognizing vehicle logos can well assist the transportation department and security department for effective, efficient and safe transportation management. As a typical target identification problem, vehicle logo recognition usually represents the process of detecting the logo area and then automatically recognizing it from static digital images or video data streams, which is mainly implemented through feature extraction and classifier construction.On the basis of extensive research on vehicle logo recognition technologies for dealing with the inadequacy in the treatment of nonlinear image features by using Principle Component Analysis(PCA) method in domestic and overseas contexts, this paper adopted the Kernel Principle Component Analysis(KPCA) to extract features from vehicle logo images. KPCA can keep parts of the nonlinear features of vehicle logo images and eliminate redundant image information as much as possible at the same time, which can ensure adequate and effective feature information in the process that image data is mapped from a high dimensional space to a lower one. The above property has been verified through simulation experiments on a vehicle logo database gained under complex environment. Experimental results show that KPCA performs better than PCA in feature extraction when they are carried out on the Minimum Distance Classifier(MDC) and Support Vector Machine(SVM) respectively.Besides, considering that widely used vehicle logo recognition methods have some defects, for instance, there are too many super parameters; it is difficult to determine those parameters and the prediction results are not ideal, Gaussian Process Classifier(GPC), of which the appropriate parameter is easy to be found and the learning capability is strong, is proposed to recognize vehicle logo. Firstly, image pre-processing methods are used to reduce interference that is caused by complex environment to image data, then KPCA is used to fully extract out vehicle-logo’s linear and nonlinear feature information. Finally, feature information is respectively entered into Probability Gaussian Process Classifier(P-GPC) and Least Risk Gaussian Process Classifier(LR-GPC) for recognition.Simulation results show that ideal effect and strong robustness were gained by using GPC methods to recognize vehicle logos, among which, polynomial kernel KPCA combined with LR-GPC obtained a relatively satisfactory classification effect and recognition speed, which proved that it is a practical new method for real-time vehicle logo recognition.