License Plate Location And Recongnition Based On Deep Learning

Intelligent transportation systems are widely used in transportation,surveillance and security industries,and license plate recognition algorithms are one of its core technologies.Traditional license plate recognition algorithms generally extract license plate features manually to locate and recognize the license plate.However,in complex environments,the recognition accuracy of traditional algorithms is not high enough and the robustness is poor.In order to improve the accuracy of license plate location in complex environments,this paper optimizes the YOLOv3 algorithm based on deep learning theory and adopts an end-to-end multi-label classification method.The main research contents is as follows:(1)Research on License Plate Location AlgorithmTraditional license plate location algorithms are easily affected by changes in external factors such as light intensity,resulting in location failure or inaccurate locating.This article adopts the target detection algorithm in deep learning to locate the license plate.In order to improve the location accuracy,a feature pyramid structure is designed and the YOLOv3 algorithm is optimized.Using the license plate location test set to evaluate the algorithm,the results show that the optimized location accuracy can reach 99.47%.(2)Research on the Algorithm of License Plate Character RecognitionTraditional character recognition algorithms usually recognize the license plate characters after segmentation.Character segmentation is greatly affected by environment factors,and it is easy to cause problems such as inaccurate segmentation,which affects the accuracy of character recognition.In this paper,an end-to-end multi-label classification method based on deep learning is used to design a convolutional neural network for license plate character recognition.Seven fully connected layers are connected in parallel at the end of the network,and the character category at the corresponding position is predicted.In order to prevent overfitting when training the model,the designed network is optimized by randomly discarding neurons and batch normalization.The performance of the license plate character recognition algorithm is tested through experiments.The results show that compared with the traditional license plate character recognition algorithm,the algorithm improves the accuracy and speed of the license plate character recognition.(3)Research on License Plate Recognition AlgorithmCombine the proposed license plate location and character recognition algorithm to form a complete license plate recognition algorithm.In order to verify the effectiveness of the algorithm,experiments are carried out using pictures with complex backgrounds.The results show that the license plate recognition algorithm proposed in this paper is able to run stably in a complex environment.The recognition accuracy is 94.67%and the recognition speed is 204ms/sheet.In the video,due to the relative movement of the shooting equipment,the car and the license plate,then motion blur occurs.This paper proposes that the pictures intercepted from the video should be clarified first,and then the license plate can be located and recognized,so as to improve the accuracy of license plate recognition,and use experiments to verify its effectiveness.