Extraction Of Vehicle Information Based On Monitoring Video And Track Query System Implementation

Nowdays,with the growing number of mobile vehicles in the city,mass traffic information under large data background is a very valuable resource.Making foll use of traffic surveillance and control system's vehicle pass records,analysis of city traffic information,to obtain more abundant traffic rules and traffic flow characteristics,not only help the government to formulate effective strategies to solve the traffic problems,such as city congestion,accidents,but also alleviate traffic pressure.It is convenient for people's travel.In this paper,video information extraction of the vehicle,the information storage and information applications are discussed.From the vehicle history track point of view,firstly,we conduct research on the vehicle detection,license plate recognition and vehicle type coarse-grained identification to obtain vehicle's informations based on monitoring video,generating the vehicle pass records in monitory site and storing them into the database.Then a historical track query system based on the vehicle passing records has been built by using the Visual Studio platform and the Geographic Information System(GIS)technology.The system mentioned above gets the vehicle's through points and generates historical trajectory on the electronic map by using vehicle type and license plate information to retrieval database,so as to provide an intuitive analysis platform for traffic management department to make reasonable decisions.In this paper,a fast video vehicle detection method based on MB-LBP features of vehicle combined Adaboost classifier is proposed.This method can effectively and quickly obtain the position of the vehicle in the video frame.Based on the results of vehicle detection,the license plate recognition and coarse-grained vehicle type identification are further realized in the paper.In the part of license plate recognition,license plate contains abundant horizontal and vertical edge information.Therefore,a license plate location algorithm based on canny edge detection operator is proposed,which achieves good location effect.After obtaining the license plate,the character segmentation and character recognition are further realized.In the vehicle type classification section,this paper first introduces the traditional classification method which using HOG features combined with SVM classifier.Because the data derived from vehicle's rear view images,vehicles similarity between different types are relatively high.Traditional feature descriptors can not effectively describe the high-order characteristics of images,which leading to the bad classification results.Therefore,a novel vehicle type classification method based on convolution neural network is proposed in this paper,the vehicle type classification network is designed and the recognition accuracy is improved.At the end of the paper,a vehicle track query system based on the traffic information is realized.In the condition of missing record of the vehicle in some monitory sites,the two track reappearing schemes based on the least driving cost(the shortest path)and the highest level of traffic smoothness(the least intersection has been passed)are presented,which can generate reasonable information for the driving trajectory.