Research On Vehicle License Plate Recognition Based On Neural Network

As a frontier subject in the field of transportation,Intelligent Transportation Systems(ITS)has always attracted the attention of many researchers.Adopting advanced information processing technology,communication technology,sensor technology,automatic control technology and computer technology can effectively plan and manage the entire transportation system so as to achieve a balance between roads,vehicles and users.In this system,the vehicle as the core research target,has a unique identification-license plate number,and all the information related to the vehicle are indexed by the license plate number issued by the transport department,so automatic identification plate number becomes One of the key technologies in ITS.However,the current license plate recognition system has some defects in accuracy,real-time,adaptability and other performance,so there is still some room for improvement.Aiming at some problems existing in the current license plate recognition system,this paper starts with several steps of the license plate recognition system and reexamines the performance optimization methods of license plate location,character segmentation and character recognition.The work in this paper is mainly divided into three aspects:Firstly,some characteristics of vehicle license plate are analyzed,including color,edge,spectrum,texture and so on.The edge feature information of license plate is chosen to start with.The traditional Canny operator edge detection method is improved,and an adaptive double threshold Selection and median filtering method,the improved method can more accurately detect useful license plate information,filter out unnecessary interference information.And the morphology of the license plate has been made to further remove the geometric elements of the license plate with a larger gap between the background elements,and ultimately get its exact location.2.As the image acquisition equipment and the license plate have a certain angle,resulting in license plate image tilt phenomenon,which identifies the follow-up characters is not good,so the first license plate tilt correction,and then use the prior knowledge and vertical projection method to get rid of License plate upper and lower borders and rivets,the final license plate character is divided into a single character.3.Because of the Chinese characters in the license plate,and Chinese characters with rich strokes texture features,and alphanumeric large difference,not suitable for common network features,so the use of Gabor feature extraction method of Chinese characters to obtain the direction of its strokes,And a BP neural network is constructed,in which the learning rate setting method is improved so that it can adapt to the speed of sample learning autonomously.After experiments show that the network can achieve better recognition effect.