Vehicle Recognition Based On Fusion Feature

With the rapid socio-economic development, the vehicle plays an important role in people's lives. People are increasingly dependent on the vehicle and vehicle ownership growth to an annual rate of over 10%, which caused huge urban development problems. These problems need by the ITS(Intelligent Transportation Systems) solutions, ITS provides a convenient and efficient urban transport management, in order to alleviate the problem caused by vehicles. Vehicle recognition is one of the core functions of ITS, which has many various methods and has flaw to improved.This paper put forward the face components' features of vehicle, which means extract different feature from each components. Different from the traditional local feature extraction, face components' feature of vehicle has a good reflection of the details inside the vehicle, it focuses on discriminatory performance of the different face components of vehicle.it means inherent differences between local features are no longer ignored as traditional method. Face components' feature of vehicle include lamp component's edge feature, grid component's texture feature, bumper component's SURF features. They respectively perform the components' most prominent vision information.This article uses bag of words model to integrate different face components' features into a global vector can be characterized vehicles. For lamp, bumper components, to build their own visual dictionary, image's visual bag of words constructed from the visual dictionary, feature vector access ing by word frequency histogram built by bag of words. Texture feature vector obtained by the response of the Gabor filter. The response of different directions of the Gabor filter determine the main direction of the grid, the total response value and the extreme response value number, response value distribution. Then combine these data as a texture feature vector. Fusion vector obtained by cascaded three component feature vectors. fusion feature vector's dimension is the sum of three feature vectors' dimension.In this article used vehicle recognition process, for input image of the vehicle, using the vehicle detector to determine the position of the vehicle, use horizontal, vertical integral projection to maximize vehicle face. Further pre-correction by eliminated license plate useless information and denoising. Component feature extract form Maximizing vehicle face to obtain fusion feature vector by bag of words model, then match the data in the database to find sample of optimal vehicle.Experimental results show that the method is effective and feasible, good stability, particularly a larger increase in time efficiency. it meet the real-time requirements of vehicle recognition system.