Research And Implementation Of OpenCV-Based Vehicle License Plate Recognition

With the gradual popularization of the car, traffic congestion often occurs, and the Intelligent Transportation System is becoming the hot research area which people pay close attention to. The Vehicle License Plate Recognition(VLPR) is an important part of the intelligent transportation system. With VLPR, the vehicle license number can be identified in real time, which then improve the automatic level of vehicle monitoring and management. Due to the complexity of the application scene, there are many shortages of VLPR, such as low accuracy of license plate locating, inaccurate character segmentation and the recognition error of inclined license plate. How to improve the recognition accuracy of the VLPR become the focus of attention. In this paper, in order to achieve the license plate recognition, license plate location, character segmentation and character recognition have been deeply discussed. The main contributions of the paper include the following aspects:(1) A license plate location algorithm based on color feature and projection information is proposed. When it has many similar color noise area in the background, the locating effect of traditional vehicle license plate location algorithm is poor. It’s difficult to choose a projection threshold in the license plate location algorithm based on projection information and too large or too small of the projection threshold can affect the final locating accuracy. As the license plate regions generally have a fixed color feature, using the characteristic of HSV color space distinguishing different colors, plate background region judgment is done in the HSV color space were, so as to obtain more accurate background license plate region. Due to the obvious characteristics of the license plate region projection, it’s easier to choose an projection threshold in row location and column location than in traditional projection method. At the same time, due to the candidate area judgement in row location, more candidate license plate is obtained, the correct rate of the license plate is improved.(2) A character segmentation method using vertical projection is proposed. The traditional vertical projection character segmentation method uses the same projection threshold for all characters, so the accuracy rate of Chinese characters segmentation is low. As the Chinese characters strokes is more, its vertical projection is generally greater than that of the non Chinese characters. This paper uses different vertical projection threshold in the segmentation of Chinese and non Chinese character, then it has higher segmentation accuracy of Chinese characters.(3) A character recognition algorithm based on character feature is proposed. One or several of the traditional character recognition algorithms usually select grid feature, feature in the horizontal direction and vertical direction feature as the character features, which can not accurately reflect the character characteristics. The character recognition algorithm which is proposed by the paper choose this three characteristics and the number of white pixel in character images as the character features, so that the selected character features can reflect both local details of the character and the overall character shape distribution.The structure of Chinese characters is more complex than that of English letters and numbers. The paper designs the Chinese characters feature library and alphanumeric characters feature library, with which the performance of Chinese characters recognition is improved.To validate the theoretical results of proposed in this paper, the paper design and realize an VLPR system. And many relevant experiments were carried out with the vehicle image in different light conditions. Experimental results show that the proposed algorithm can be accurate in the license plate location and character segmentation, and the correct rate of the character recognition is also higher, and it has a certain practical value.