Research On License Plate Recognition System In The Intelligent Park

With the rapid increase of city vehicles, intelligent park has become the research hotspot of intelligent building.Vehicle license plate recognition system (LPRS) is the core of the intelligent park, consists of four main modules, which are license plate location, tilt correction, character segmentation and character recognition. This paper carries out systematic and comprehensive research on the key technology after summarizing a lot of latest materials at home and abroad in recent years.In the stage of license plate location, an algorithm of license plate location based on nonsubsampled Contourlet transform is presented. The algorithm can locate the license plate accurately. Firstly, the license plate picture is processed with nonsubsampled Contourlet transform to get 8 oriented high-frequency components pictures of the original picture. Secondly, these high-frequency components pictures are merged into one high-frequency components picture which can highlight the plate through a certain combination rule. Finally, locate the license plate with mathematical morphology and connected components analysis.The tilt correction stage, two plate horizontal tilt correction methods based on the minimum variance of edge projection and on K-L expansion respectively, and a plate vertical tilt correction method based on the minimum variance of projection are proposed. They are compared with the commonly used methods of plate tilt correction. The experimental results show that the proposed methods have higher precision, better robust, real-time-oriented.In the stage of character segmentation, the dot in the middle of the plate is removed firstly. Then, each character segmentation position is identified from the position of the dot to left and right respectively according to the vertical projection message of the image. In addition, some simple and effective treatment is done to fault or adhesion character.In character recognition, the Krawtchouk moment invariants and trace transform methods is chosen to extract character features, and SVM classification is used as the method to identify characters. The experiments show good recognition results and real-time-oriented.Finally, the license plate recognition system is realized with programming. A lot of experiments to actual accessed plate images show satisfactory results. An intelligent park management system is built on the license plate recognition system, and the principle, processes and functions of the system are introduced.