Research On Location And Recognition Of License Plate With The Method Of Harris Corner Detection And AP Clustering

Intelligent Transportation has begun to be applied to parking fees?highway fees?traffic data collection?access control?crime prevention and other traffic areas andsecurity. Intelligent transportation system will be the main trend of traffic managementsystems in the near future. The license plate recognition system is the intelligenttransportation system indispensable component. License plate recognition systemgenerally consists of license plate image preprocessing and license plate location?license plate character segmentation and character recognition. This article focuses on thetechnology of the license plate recognition under complex backgrounds.1) License plate image preprocessing. In this paper the methods we use consist ofgray image processing?smoothing filtering of gray images?edge detection of gray imageand morphological processing.2) Plates location. Considering the license plate area contains more information andcorner edge information?this paper presents a method containing Harris corner detectionalgorithm and AP clustering algorithm. First?Harris corner detection algorithm figure outthe license plate area?and then we eliminate outliers by AP cluster.Finally we merge theneighboring area and recognize license plate area by the ratio of the width and the height.3) Character segmentation.License plate character segmentation is particularlycritical in the whole process? segmentation accurate or not?will determine whether thecorrect license plate character recognition. Firstly?we correct the license plate area byHough transform and projection?and then remove the borders by gray projectionhopping?finally we segment characters by means of projection and template matching.4) Character recognition. We identify the individual characters based on priorknowledge and study the BP neural network. We optimize the BP algorithm by adaptivelearning rate adjustment and additional momentum. And then according to extractedusing template matching methods?we train and identify the splitted letters and numbersby means of BP neural network after feature abstract from the samples of the numbersand the letters.In this paper we choose Matlab as the research tool to study vehicle license platepositioning?character segmentation and character recognition in VLPR.Experimentsshow that a robust image preprocessing will help a lot in vehicle license platepositioning.Under the premise of proper image preprocessing,the methods we proposecombined by Harris corner detection and AP clustering will get high accuracy.Themethod combined by projection and template matching does well in charactersegmentation,but the accuracy of VLPR needs further improvement.