Article Preview
Top1. Introduction
With the rapid development of intelligent transportation information system, license plate recognition technology plays a very important role. The promotion and popularization of license plate recognition technology will have a significant and far-reaching impact on urban road management, reducing the occurrence of traffic accidents and vehicle theft cases, and ensuring social stability. The whole license plate recognition system is divided into four parts: image acquisition, license plate location, character segmentation and character recognition. With the improvement of computer performance and the development of computer vision technology, license plate automatic recognition technology is more and more mature, and the system is more and more perfect. However, there are still some problems in practical application. For example, what kind of preprocessing technology is used before license plate character recognition, which is more conducive to the subsequent license plate positioning, character segmentation and character recognition, and how to deal with license plate tilt, and what kind of license plate location method and license plate character segmentation technology can improve the accuracy of location and segmentation and the operation efficiency of the whole system. In the paper, based on the theory of existing algorithms, key technologies before license plate character recognition are studied in detail, such as vehicle image gray scale, binarization, edge detection, median filtering, license plate positioning and tilt correction, as well as license plate character segmentation algorithm.
In view of the complex background, Jing Zhang et al. 2017 proposed a new method to detect vehicle license plate. Firstly, the image is transformed from RGB color space to YUV color space. Because the license plate is usually in a certain range near the tail lamp, the V component is used to detect the rear lamp, and then the candidate area of the license plate is determined according to the prior knowledge. Then, in this region, Sobel operator is used to obtain the vertical edge of the image, and a new edge energy map method is used to process, and two points are determined in the vertical and horizontal directions respectively to detect the license plate as the rectangular vertex. The experimental results show that the accuracy of the algorithm is 93% and the operation time is about 0.21 seconds. It can overcome the influence of complex background and achieve real-time and accurate performance, but the accuracy and operation efficiency are not satisfactory.
In order to locate the license plate in the image quickly and accurately, Weizhao Zhong et al. 2017 proposed a license plate location algorithm based on the extraction of character edge points of license plate. Sobel operator is used to detect the edge of the image, and a fast color feature extraction algorithm based on the license plate features is adopted. In order to eliminate other interference edge points, an edge point window scanning algorithm based on the proportion of pixel points around the characters is designed, and advanced morphological transformation is used to extract the candidate area of the license plate. The experimental results show that the success rate of this method is 98%, which is applied to the vehicle body and the bottom of the license plate under the same color, so that the license plate can be located successfully.
Yuke Wei, Fa Ouyang 2018 proposed a license plate location algorithm based on the difference value of adjacent pixels. In the pre-processing stage, the circular arc function is used to process, and the difference value algorithm of adjacent pixels is proposed to extract the edge of license plate characters; edge detection, block processing, and high edge block reorganization are carried out to screen out candidate areas; in the candidate confirmation stage, the R+G channel graying method and adaptive mean binary method are proposed to process candidate areas, Combined with the character distribution characteristics of license plate, the pseudo candidate area is filtered and the license plate area is obtained. The experimental results show that the algorithm has a high positioning accuracy of 97.5% and a running time of 0.62s. It has a certain practical value, but the running time is a little longer.
In order to reduce the impact of shooting distance and light on license plate recognition and improve the accuracy of license plate recognition in complex background, Yan Wang et al. 2019 proposed a new license plate detection and recognition method based on the maximum stable extreme regions “MSER” and stroke width transform “SWT”. The experimental results show that the accuracy of this method is 94.86%.