Research On License Plate Recognition Technology Based On Consecutive Frames Difference

The Location and Recongnition of License Plate is one important topic of the Intelligence Traffic Management. License plate recognition (LPR) has great application prospective in tolling, monitoring traffic flow, investigating on-the-spot traffic accident, handling breaks of the traffic regulations and keeping parking secure. People pay increasing attention to LPR based on digital image processing and pattern recognition.Rearch on key technologies of LPR have been done in this thesis, including preprocessing, locating and recognizing. A LPR laboratorial system which can recognize the plate in natural video is designed and implemented.The current license plate location technology mainly consists of location technology based on grayscale image and location technology based on color image. Frames difference-based license plate location technology is researched in this thesis. The rectangular (window) search method is modified, which improve the efficiency of searching the plate. Frames difference-based , grayscale image-based and color image-based location technology are used in this thesis,got good performance. In character segment, a method of direct averagely segment to some widthly segmented character image is researched. And another accurate locate for every single character image is carried out. In the size normalization of character image, a method is presented: normalize the the character to 40 pixels by height, then copy the result to a black image of 40*40 pixels size. In Character Recognization, the features presented by Peter W.Frey and David J.Slate is used, and the Radom Forest presented by Leo Breiman is used as classifier. The result of compared with several other classifers, show that the approach perform well in efficiency and reliability in recognition.