A License Plate Recognition Algorithm Based On Gradient Template Features

With the dramatic development of economy, the number of motor vehicle is growing at a high speed. However, at the field of transportation, the speed of infrastructure construction is still far behind the rapid growth of the number of automobiles, which caused a series of social problems. Traffic congestion has become a common phenomenon in major cities. In order to improve the utilization of transportation resources, improve transportation efficiency, reduce traffic congestion and reduce energy consumption, the Intelligent Transportation System (ITS) is received a wide attention and is growing rapidly.As an important concept in Intelligent Transportation System, the License Plate Recognition (LPR) technology is a major task on the implementation of automatic traffic monitoring and management. It can provide important data for the traffic monitoring and management,therefore, it is considered as a significant means of information acquisition. From a functional perspective, License Plate Recognition System can be generally divided into three major parts: 1. License plate locating, 2. Character recognition, 3. Related image processing works. In the research of license plate locating, this paper proposed a license plate precise locating and segmentation algorithm that based on gradient template features. Focusing on the general structure of the license plate area, this algorithm can achieve a high locating accuracy. This paper also analyzed the application of the Cascade algorithm in license plate locating, thus constitutes a high positioning accuracy and efficiency License plate location classifier. In the research of character recognition, this paper proposed a license plate tilt correction algorithm based on the growth of connected pixels. This paper also analyzed the use of united character features that are based on the statistical features and geometric features of the license plate character.The experiments show that the proposed license plate locating algorithm has a relatively high accuracy and reliability. The success rate of coarse positioning is 95.4%, and the success rate of precise positioning is 92.3% and the positioning speed is 300 to 500 ms per image. And the algorithm can achieve a relatively high robustness when the contrast is low or the license plates are contaminated, etc. In the character recognition experiment, the overall license plate recognition rate was 93.59%, and the character recognition rate was 98.47%, which reached a satisfactory level.