A Study Of Vehicle Recognition Technology Based On Gabor Features And Sparse Representation

Vehicle recognition technology is an important research field of intelligent transportation system, is also a challenging issue in the fields of image processing, machine learning, computer vision etc.. This technology can be applied to highway management system, traffic monitoring and vehicle detection systems, vehicle management system, vehicle payment system and other areas. This technology has high practical value in regulating urban transport security, fighting against vehicle theft, preventing and reducing traffic accidents, etc..In this paper, firstly the current status of vehicle recognition technology is introduced briefly, some of the challenges and issues, with which vehicle recognition technology is faced, are analyzed and summarized. The classic method takes the global features of samples into account, have a high computational complexity and poor robustness to changes in scale, illumination and other conditions. Gabor features can describe the image local feature information on different scales and in different directions, and can tolerate changes in the illumination, scale, etc. This paper draws on the sparse representation classification algorithm in face recognition, a method of vehicle recognition technology based on Gabor Features and sparse representation is proposed.The implementation of vehicle recognition based on Gabor Features and sparse representation mainly includes three modules:the establishment of sample database, feature extraction and vehicle classification. In this paper, the vehicle image samples are classified firstly, and then preprocessed, obtained the required vehicle face image samples, a total of 60 different types of vehicles are selected to recognition, each type has 10 pieces of vehicle face image, which changes in illumination, the image size is 128 * 64; then use the Gabor filter to extract a sample image feature information; and finally use the classification method based on sparse representation to determine the testing sample belongs to which category. Firstly, doing experiments respectively on six different sizes of data sets (10,20,30,40,50,60), making full use of the local features of the samples, to verify the feasibility of the proposed algorithm in this paper. The experimental results show that the proposed algorithm is feasible, the average recognition rate can reach 92.8%, and has strong robustness to illumination changes. Then by comparing the recognition results of the sparse representation classification algorithm and the algorithm proposed in this paper verify the effectiveness of the proposed algorithm. The disadvantage is that the proposed algorithm improves the recognition and enhances the robustness, while paying the price of time.