Research And Realization On The License Plate Recognition System

At present in China, with improvement of people’s material life, vehicles become moreand more popular and we have to face increasingly heavy traffic pressure, how to manage therapidly increasing vehicles effectively cannot be ignored. In this context, intelligenttransportation systems (ITS) emerged, as the critical technique, the vehicle license platerecognition (LPR) system occupies important place in road pricing, vehicle parking andtraffic regulation such as project management.The complete vehicle license plate system mainly consists of three parts, the vehiclelicense plate location, character segmentation and character recognition. So the three separatealgorithms are studied in this paper:Firstly, considering our country’s license plate contains Chinese characters and Chinesecharacters have rich texture features characteristic, a license plate location algorithm based onSIFT algorithm is proposed, a standard character library is created, the approach extractsSIFT features of the license plate in the sample template database, matches the SIFT featuresof the license plate image with them, then eliminates the error matching points with RANSACalgorithm and calculates the affine transformation matrix of the template, thus the approachachieves accurate initial license plate location and initial tilt correction. Then transform theextracted license plate region to binary image, obtain the incline angle with the Radontransform to calculate the accurate affine transformation matrix, finally achieve the precisevehicle license plate location and tilt correction. Experiment shows that compared withtraditional algorithms, the proposed method not only achieves accurate license plate locationand tilt correction,but also has good adaptability to the image brightness, defilement, tilt,scale change and so on, it has a good performance on robustness.Then, this paper proposes a license plate character segmentation method which unitesprojection and template matching,the located license plate area has to be processed by usingmathematical morphology for de-noising texture enhancement, and use projection method toremove the upper and lower, left and right borders, segment characters by using our country’slicense plate character spacing feature. The results of experiment show that the method cansplit off the license plate characters accurately, and has good adaptation to character blocking,adhesion and crack condition. Due to the Chinese character has been identified in the plate location part, it is only toidentify the remaining six alphanumeric characters. Firstly, reduce and normalize thesame-sized characters; And then extract the next six characters’ sub-block barycenter feathers;Finally each characters are trained and predicted by SVM with RBF kernel function bycomparing the other two. One-Against-One classification method is adopted in this step.Experiments show that the method presented in this paper has higher recognition rate in thecase of small-samples.