The Research And Implementation Of Vehicle Type Recognition Based On Video Sequences

With the rapid development and improvement of the national economy and people’sstandard of living, traffic increasingly environment problem and traffic management issueshave become more serious, consequently, Intelligent Transportation Systems(ITS) using theexisting road transport facilities and various high-tech science technology to solve trafficproblems has achieved great development.Vehicle Type Recognition (VTR) is the main task and the key technology of ITS, butmore types recognition accuracy of VTR based on video sequences under complexbackground is not high, in order to solve this problem, firstly, self-Adaptive backgroundupdate algorithm is brought into background image difference method in this paper, it extractscomplete moving target from complex background and overcomes disadvantages ofconventional moving target extracting method.Secondly, vehicle geometric features are extracted based on label algorithm ofconnecting area and boundary tracking theory, a feature extraction method is proposed byconstructing local information entropy matrix of vehicle face, which reflects the gray andspatial texture information, so a combination features of vehicle type are extracted, it meet therequirement of extracting vehicle combination features and avoids ineffectiveness of thesingle feature for recognition system, for the good performance of Gabor function in digitalimage analysis, a vehicle features extraction method with Gabor filter is proposed, whichextracts the maximal energy value of vehicle after Gabor filtering as vehicle feature, and theenergy matrix search process is gave, it achieves the purpose of extracting effective featurewith Gabor filter and providing basis for correctly recognizing vehicle type.At last, a recognition method for vehicle combination features is proposed based onvehicle geometric features recognition with three-layer BP network and matching recognitionof vehicle face feature, and a improved BP network is used to recognize vehicle Gabor feature,analysis of experimental results show that the proposed methods have higher accuracy rateand achieve the purpose of recognizing8vehicle types, it is applicable to a VTR system withreal-time visual surveillance and has theoretical research and practical applications value.