Research And Implementation Of Vehicle Classification Method In Surveillance Scene

In recent years, the application of intelligent transportation system is more and more extensive. Vehicle classification system is an important branch in the application of intelligent transportation systems. With the development of technology such as computer vision and image processing, the vehicle classification system in the surveillance scene has been paid more and more attention. However, due to the complex background of the surveillance scene,it is susceptible to various weather conditions, including light, rain, snow, etc., while with the diversification of vehicle models, vehicle classification faces more challenges.In this thesis we study the vehicle classification method under the surveillance scene. Firstly,we propose a algorithm based on the combination of moving object foreground extraction and deep learning based on object detection, which can obtain the vehicle target in the complete and non-adhesion surveillance scene. Secondly, In this thesis, using deep learning network, through training to obtain vehicle classification model, and constructs a rich data set, which contains 11 vehicle categories, 6 driving directions, and uses the data set to train the vehicle classification network model. Thirdly,we propose a vehicle-type correction algorithm based on regional tracking method to improve the accuracy of vehicle classification . Finally,based on the above algorithm, we design and implement the vehicle classification system under the monitoring scene,and this system is tested in a real-world scenario.The test results show that the system has good robustness in vehicle classification.