The Design And Implementation Of Vehicle Recognition System Based On Video Image

Vehicle recognition is an important part of the Intelligent Transportation System, and vehicle recognition may play an important role in the traffic control, automatic parking collection system and highway toll stations. Vehicle Recognition based on video sequences has some characteristics, such as intuitive, economy and rich information, therefore the research of vehicle recognition based on video sequences becomes more meaningful and has rapidly developed. Meanwhile, with the continuous improvement of the social demand, the accurate and timely vehicle recognition system will have a great social demandIn this thesis, we designed and implemented a vehicle identification system based on video sequences, which was on the base of study and comparison of domestic and foreign intelligent transportation applications. In this thesis, we discussed the following problems:Firstly, we designed a video-based image recognition system models, built the hardware platform of the recognition system with cameras and PC. The hardware platform can help to obtain the real-time vehicle video on the road, which was used to get the video image information for vehicle recognition. As to the software of this system, we designed and implemented an identification method based on multi-feature fusion, according to the comparative and analysis of existing recognition algorithm.Second part is video image pre-processing and vehicle detection. The video image pre-processing was implemented by graying and filtering the video image captured by cameras from the road,aiming at removing noise and reducing the irrelevant information of video image.In this thesis, we achieved a target detection method based on background subtraction Gaussian mixture model which had achieved an effectively detect in the application scenarios in this article.Third part is vehicle feature extraction and vehicle tracking. To extract the vehicle feature to achieve the vehicle classification with the vehicle test results above. In this thesis, we applied a multi-feature fusion method including vehicle geometric characteristics, histogram of Oriented Gradient and speeded Up Robust Features. This method can more fully describe the characteristics of different types of vehicles; and the accuracy of vehicle recognition can be greatly improved. To further enhance the accuracy of vehicle identification, in this thesis we selected the Kalman filter tracking method for vehicle tracking after analyzing two different tracking methods. Recognize the same vehicle multiple times with vehicle tracking, and then analyze multiple recognition results to get the final classification results, reduce the single classification error caused by misclassification in identification process, and improve the accuracy of recognition eventually.Lastly, the realization of vehicle classification and over all test. The SVM classifier the system needed was achieved, based on the analysis of the source code of SVM support vector machine and LibSVM. Use the existing training samples to train the SVM classifier, through continuous testing and modification of the vehicle classification parameters to achieve better classification results in the training process. Finally, implemented the system performance test of vehicle Recognition with road vehicles-Video to verify the method proposed in this thesis, and analyzed the final test results.