Research On License Plate Recognition Algorithm In Haze Weather

Smart transportation provides a solution to urban traffic problems and makes urban supervision more efficient.As the core technology of intelligent transportation system,license plate recognition is widely used in various traffic scenes.Affected by many factors such as economic development,China’s major cities are prone to haze weather,which leads to a significant reduction in the accuracy of license plate recognition,and brings great challenges to the intelligent transportation system.To solve the above problems,this paper combines license plate recognition technology with image dehazing technology,and proposes a license plate recognition algorithm in haze weatherIn image dehazing: in order to avoid the color contrast distortion of traditional image dehazing algorithm,this paper proposes an image dehazing algorithm based on recursive residual network.This algorithm takes different scales of hazy images as input,and extracts deep features through recursive residual network to avoid gradient vanishing.Finally,the algorithm overcomes the uncertainty of atmospheric scattering information and the direct fusion of different scales.In the process of information transmission,we add different weights to different information.The results show that the PNSR and SSIM values of this algorithm are 25.46 and 0.9029 respectively in the synthetic hazy image,and the IE and AG values are 26.36 and 8.1358 respectively in the real hazy image.In the aspect of license plate recognition: in order to improve the ability of license plate recognition,this paper proposes an end-to-end license plate recognition algorithm based on YOLOv3.Firstly,this algorithm constructs a new license plate data set and labels the license plate and characters in the data set;secondly,based on the model of yoov3,the structure of darknet-31 is proposed,which not only improves the extraction ability of the network,but also speeds up the extraction speed;aiming at the characteristics of small license plate characters,a network prediction scale is added to improve the detection ability of license plate characters.Compared with the traditional YOLOv3 algorithm,the average accuracy of this algorithm is improved by 4.1%,the recall rate is improved by 3.8%,and the speed of character recognition is improved by20%.The experimental results show that this algorithm has better robustness and real-time performance.Finally,the image dehazing algorithm and license plate recognition algorithm in this paper are combined for experimental research,and the samples of correct license plate recognition and wrong license plate recognition in haze weather are analyzed to further verify the effectiveness of the algorithm.