License Plate Recognition Algorithm Research And Application

License plate recognition system(LPRS)is one of the most important parts of intelligent transportation system.Traditional LPRS has achieved a great accuracy in some particular scenes,but it still exists some problems in some scenarios,for LPRS can be affected by many other circumstance factors.Now the fast development of Artificial Intelligence and Internet of Things put forward higher request to LPRS.This thesis comes up with some new improvements about the process of LPRS and some improved algorithms related,especially in license plate location,character segmentation and character recognition.First,the background and research status of LPRS are introduced.Then the components of LPRS are addressed.After that,we introduce some regular plate location algorithms,and use Open CV to process the photos which contained license plates.After the image denoising,edge detection,image threshold and morphological operation,the license plate can be located and segmented precisely.Second,we implement a character segmentation based on histogram statistic on binary images.Our results indicate that this method is accurate and effective.Then we study some regular image features,and construct a license plate character database based on features of image gray values and histogram of oriented gradient(HOG).This database contains more than 5000 training photographs and over 1700 testing photographs.Besides,we study the principal component analysis(PCA)algorithm,and use it to decrease dimensions on our character images,and to judge the car's running main axis.Finally,we study some regular character recognition algorithms.K nearest neighbor algorithm which is based on image similarity and support vector machine(SVM)algorithm are adopted to recognize the plate character and achieve a high accuracy.Then we propose a character recognition algorithm based on Classification and Regression Trees to recognize plate character.Compared with the SVM algorithm,K nearest neighbor algorithm and Random Forest algorithm,our algorithm can achieve nearly the same accuracy,and be more effective and easy.