Research On Technology Of License Plate Location And License Plate Character Recognition

With the development of national economy, the number of vehicles increases rapidly. Although makes our life convenient, it causes a lot of traffic problems. Intelligent traffic surveillance (ITS) has been the main measure and development direction of high way traffic and municipal traffic management today, and the license plate recognition (LPR) technique is the kernel of ITS. A typical LPR system consists of three parts:license plate location, license plate character segmentation and license plate character recognition. This paper concentrates on license plate location and license plate character recognition.I. License plate location. Aiming at the interferes of the illumination during the imaging process and the defile on the license plate, propose a license plate location method based on machine learning. Firstly, we convert the color license plate image into grayscale image, then enhance and denoise it; Secondly, we employ Harr-Like feature-based Adaboost cascade classifiers to obtain candidate rectangle license plate regions; Finally, the support vector machine with posterior probability output is applied to test the candidate regions and decide the final results. The experimental results show that the proposed algorithm is feasible, robust and applicable to locate license plates, whether the detected images are complex or not.Ⅱ. License plate character recognition. A new algorithm is proposed for gray-scale image license plate characters recognition. At first present four license plate character Local Binary Pattern (LBP) operators, and apply them to calculate LBP texture images in four directions. Then, two-stage blocking approach and histogram are applied to every LBP texture images to obtain feature vectors, which are used to construct support vector machine classifier. Finally, the recognition result confidences are obtained by analyzing the posterior probability output of Support Vector Machine (SVM). The experimental results show that the algorithm for gray-scale character image license plate recognition has strong robustness and good practicality.