Design Of Classifier Of License Plate Location And Relationg Algorithms Research

License plate location is a core module for intelligent transportation systems, but it is verysensitive to illumination, angle changing, the quality of images and so on. These difficulties limitlicense plate location's application area. They are also the problems which the thesis studied.Some new development about image-processing and recognition technology are analyzed in thispaper, and then an algorithm about license plate location which is robust in complexenvironment and reduces the difficulties in project deployment is described. It is also helpful toother object detection problems.The proposed algorithm includes two parts. In the first part, some new Haar-like features forChinese license plates are proposed. Then, the algorithm uses an ensemble classifier, containingmany AdaBoost classifiers, which are trained by different features and examples, to detectlicense plates in an input image, and finds more than one plate candidate regions which look likeplates. Such regions, located by the ensemble classifier, built a plate candidate set. In order toimprove the classifier's volume, the algorithm uses Intel threading building blocks to change theserial program for parallel one. In the second part, the algorithm uses a Real AdaBoost classifierwhich is trained by HOG to give each plate a confidence-rated, which can indicate thedifferences between the candidate regions and the real license plates. Then these candidates aresorted by their confidence-rated. The candidate region, which has the largest confidence-rated, isconsidered as a license plate.Tested in a large test set, including 27018 license plate images, the hit rate is 96.2%. Thealgorithm, which has a low requirement to environment changing, and whose volume can meetthe challenge in generally scene, can be used in park, crossing and so on.