On Vehicle Recognition Based On Gabor Features Sparse Representation

Intelligent transportation system(ITS) is a critical component of modern traffic, with the development of intelligent transportation as the key areas in computer vision and intelligent transport systems, vehicle recognition technology is also increasing attention of researchers. The technology in the field of traffic monitoring, highway systems, vehicle detection management systems, electronic payment systems and automated parking building has a wide range of applications. It has outstanding value in alleviating pressure on urban traffic, urban security and reduce traffic accidents and so on.The vehicle recognition based on Gabor sparse and sparse representation founded by establish vehicle sample database, Gabor feature extraction and vehicle classification algorithm. This paper selected 60 different types of vehicle type to identify, each of the model composed of 10 different cars face images, which changes in angle and illumination. After use these images to establish the sample library. Then use Gabor filter to collect sample image feature information, at last use sparse representation classification method and its improved method to test image classification. This paper uses three classification algorithms during the experiment, they are SRC,GSRC and based on VTV Improved GSRC. We use these three classification algorithms do comparative test on different sizes of sample library, the test results demonstrate the feasibility of the proposed algorithm. Compare the experimental data of three different algorithms,GSRC contrast to SRC, recognition rate increase by 31%, with an average recognition rate of 90.9%.Then use VTV instead of the step of the original method to solve the minimum 1l norm propose an improved algorithm. Making the recognition rate model has been improved to reach 93%, especially in the case of low-dimensional significantly enhance the recognition rate. Experimental results show that the vehicle identification method takes full advantage of local characteristics of sample images, have robustness on the changes of illumination and scale. But the disadvantage of this method is that the more time consuming,average time of identification process increased from 70.1s to 360.2s, affecting the real-time identification.