Research And Application Of License Plate Recognition Based On Deep Learning

As an important part of intelligent traffic management,license plate recognition systems are widely used in real life,such as highway toll stations,intersection traffic control,community and parking lot automatic toll release system.These specific applications can effectively alleviate traffic congestion,save manpower and time costs,and improve work efficiency.At present,research on license plate recognition technology for specific environments has been relatively mature.As the main steps of the traditional license plate recognition algorithm,license plate location,character segmentation and character recognition,the obvious disadvantage is that the accumulation of errors between modules is likely to lead to a lower accuracy of the final license plate recognition.Especially in complex situations such as insufficient lighting or blurred license plates,errors between modules can seriously affect the accuracy of the final license plate character recognition.Therefore,research on license plate recognition technology in complex environments still has important market value.The specific work is as follows:The local and foreign license plate recognition algorithms are reviewed.The basic neural network algorithms are analyzed,and the application of deep learning algorithms in license plate recognition technology is also analyzed in this thesis.The accuracy of license plate location has a crucial influence on the recognition rate of the license plate recognition algorithm.Based on the existing target detection algorithm,a license plate location algorithm based on deep learning is implement ed.Compared with the traditional algorithm,the algorithm in this paper does not need to perform tedious preprocessing on the input image,and the algorithm has high detection precision.When the target learning algorithm of deep learning is used for license plate location,the selected detection target is the license plate area.Similarly,the detection target can also be a single license plate character,but the area of the character is much smaller than the license plate area,which leads to a larger Detection difficulty.In this paper,the application of deep learning in license plate recognition is deeply researched.An end-to-end license plate recognition algorithm is proposed to identify the actual license plate image.It can directly give the bounding box of the predicted license plate location and identify the corresponding efficiently and accurately.License plate number.The entire network can train end-to-end,no longer through the three steps of the traditional license plate recognition algorithm,which can effectively reduce the error accumulation.In this paper,we collect hundreds of thousands of license plate images in complex environments,and produces training sets and test sets for six different environments.It is used for training and testing of license plate location algorithms and license plate recognition algorithms.The superiority of the proposed algorithm is verified by comparison between different algorithms.