Research And Implementation Of Technologies Of Multiple License Plate Detection And Recognition For Unconstrained

Recent years,the application of intelligent transport has become more and more extensive.The license plate is the unique identifier of the vehicle,and the detection and recognition of the license plate is the main content of the intelligent transport system.Traditional license plate detection and recognition technologies for constrained scene are difficult to adapt to new needs.In the unconstrained and multi-license scene,the problem of scene diversity and low image quality makes the detection and recognition of the license plate more challenging and difficult.With the development of deep learning,the detection and recognition technology of license plates has made great progress.This paper aims to solve the problem of license plate detection and recognition in unconstrained and multi-car pass scenes.For license plate detection,this paper uses a two-stage detection algorithm based on deep convolutional neural networks.First,this paper designed a candidate frame selection network for license plates.Through the feature map calculated by the deep convolutional neural network,our network generates the quality and multi-scale candidate regions for the license plate on the feature map.Then,through the regression network,the classification and position regression of candidate frames are calculated.For the license plate recognition of double-line characters,this paper proposes an end-to-end sequence recognition network based on deep learning,The network reconstructs the feature map calculated by the deep convolutional neural network,the feature map has time-series characteristics,the sequence feature map is identified by the recurrent layer and the transcription layer.This paper designs a double-recurrent transcription network for the serious imbalance between Chinese characters and English numeric characters in license plate recognition training.For the feature map of the sequence,the network calculates Chinese characters and English numeric characters through two branches.Finally,through the prior knowledge of the license plate design,the final recognition result is obtained.The experimental results show that the proposed algorithm achieves better results than the existing algorithms in multi-license plates detection and recognition tasks for unconstrained scene.