Design And Implementation Of License Plate Recognition System Based On Vehicle Detection

With the development of ITS(Intelligent Transportation System)in China,license plate recognition system,as an important part of ITS,has attracted a lot of research and is applied in various scenes.Most license plate recognition systems extract license plate information on a single vehicle image,while the license plate recognition technology of multiple moving vehicles in viedo scenes is still not perfect.License plate recognition in video scenes will easily fail because of the complex background and interferences between multiple vehicles,and the license plate recognition system will do recognition work on the same vehicle repeatly,resulting in a lot of data redundancy and a lower efficiency.Aimed at problems mentioned above,this thesis designs and implements a license plate recognition system which can be applied in viedo scenes.The main work of this thesis is as follows:(1)The application of YOLO(You Only Look Once)detection model in vehicle detection is studied.Traditional vehicle detection algorithms generally have low detection accuracy in complex scenes.Based on YOLO target detection technology,this thesis designs an algorithm of vehicle target screening and a vehicle detector to complete the extraction of vehicle target.The method avoids duplication of recognition of the same license plate,improving both detection accuracy and system performance.(2)A license plate location method combing image processing and machine learning is designed.Traditional localization methods based on image processing are generally poor in robustness,while localization methods based on machine learning are generally slow.In this thesis,we first use color feature and vertical edge feature to extract candidate regions of license plate,and then complete the location work by prior knowledge and SVM(Support Vector Machine).The method takes both accuracy and timeliness into account.(3)A character segmentation method combining vertical projection and template matching is studied.In view of shortcomings of traditional segmentation algorithms based on vertical projection and template matching in dealing with special characters and defiled license plates,this thesis first completes rough segmentation through an improved segmentation algorithm based on vertical projection,and then uses the constructed license plate template to complete the final segmentation.Then according to charactertistics of license plate characters in China,this thesis designs several SVM classifiers to complete character recognition work,and trains those classifiers by making a comprehensive extraction of HOG(Histogram of Oriented Gradient)features and uniform grid features.On the basis of the works above,a license plate recognition system based on vehicle detection is designed and implemented.Then main modules of the system are tested,and the performance of two core algorithms is tested too.The test results show that the system functions stably.To a certain extent,the system saves traffic managers' working cost and enhance the traffic management.