Key Technologies Of Transportation Vehicle Recognition Based On Image Content

Recently, with the increase of the car ownership, transportation network becomes more and more complicated, the traffic management system is facing a serious test. Based on this background, the concept of intelligent transportation systems appears and gradually becomes the development trend of the modern transportation system. As a part of the intelligent transportation systems, vehicle recognition plays an important role during auxiliary against traffic accidents, violation of escape etc. It receives more and more attention for its broad application prospect.Vehicle recognition based on image content has a few parts which mainly includes image preprocessing, vehicle detection, vehicle-logo location and recognition. Based on the researches of vehicle image preprocessing, vehicle detection, vehicle-logo location, the key research of this paper is vehicle recognition. In terms of vehicle recognition, the car logo is selected as the recognition object. In consideration of the problem that the existing vehicle logo recognition methods cannot handle the vehicle logo recognition under unsatisfactory situations, such as cast shadows, occlusions, defaced and so on, which affect the recognition effect. So an algorithm based on discriminative low-rank matrix recovery and sparse representation is proposed. First of all, discriminative low-rank matrix recovery is used to correct the unsatisfactory training samples, then it learns a low-rank projection matrix to correct the corrupted testing sample by projecting the sample onto it corresponding underlying subspaces. At last, sparse representation method is used to classify the testing sample.In consideration of the problem of the high calculation cost of the sparse representation based vehicle recognition method, a method based on dictionary optimization and regularized robust Collaborative representation is proposed. In the method HOG feature is selected to describe the vehicle logo. HOG features were extracted from training samples and testing samples, respectively, then dictionary learning method based on Fisher discrimination criterion is adopted to optimize the train samples. Finally, regularized robust Collaborative representation is used to recognitionIn order to verify the effectiveness of the algorithm, experiments were made on Medialab LPR dataset; the experimental results show that our method has a better effect.