The Research And Achievement Of A Relatively Pervasive Vehicle Plate Recognition Technology

License Plate Recognition System (License Plate Recognition, LPR) is an importantresearch topic in intelligent transportation.LPR system is a comprehensive application ofimage processing,machine vision, pattern recognition, information theory etc.LPR systemplays an important role in intelligent transportation system.LPR system generally includes three modules: license plate localization,slant correction,character segmentation, character recognition. Based on the study and summary of theprevious work in the license plate localization and segmentation.In this paper we proposed animproved algorithm for license plate localization and character segmentation, and the realizeda complete set of LPR system.In the license plate localization stage, we adopted image pyramid method to processmulti-scale image. Then applied edge detetion and binarization. after binarization we processit with mathematical morphology.And contour analysis, screening and ranking, then outputthe candidate region in descending order according to it’s similarity with the real license plate.To achieve improvement in overall efficiency of the algorithm.In the license plate slant correction stage, this paper presents a skew detection algorithmfor license plate based on Radon transform and edge features.It contens steps as follow: edgedetetion,Radon transform,finding the angle that Radon projection result has maximum meanaverage variance,horizontal and vertical correction.In the license plate character segmentation stage, this paper presents a combind-contourcharacter segmentation algorithm based on multiple thresholding. The algorithm uses avariety of threshold method,thresholding with those method respectivly,then contouranalysing.And finally put all the contour sets together and screening.Then select the qualifiedcontours as our segmentation result.Our method combined different kind of threshold method,threshold value, pretreatment, and combine the successful parts of each result,neglected thefailed parts, so that it is robust and flexible. At last, we construct the vehicle license plate recognition system with MATLABprogramming. The system includes the license plate localization, character segmentation thatproposed by this paper. Training the BP neural network model,then recognizing the characterwith it. Finally,test our system with a large number of license plate images which wereunconstrained and collected from the Internet,the result was satisfying.