License Plate Recognition Method Based On CNN Under Complex Conditions

With the progress of society and the development of science and technology,the number of cars is increasing.Too many cars make the traffic congestion in cities jam worse,so intelligent traffic management in cities is increasingly important.As an important part of intelligent traffic management,the license plate recognition has always been a hot topic.At present,under the normal conditions,the recognition accuracy of the license plate is high,but the accuracy rate of license plate recognition is low under the strong complex conditions such as strong and weak illumination,rotation distortion and motion blur.Aiming at the license plate recognition under the above complex conditions,in this paper convolution neural network is applied to license plate recognition.By adjusting the network parameters,a convolution neural network suitable for license plate recognition under complex conditions is trained.The main work is as follows:Aiming at the inaccurate location of license plate images under complex conditions,a license plate location method based on multi feature fusion is proposed.The processed image is used for license plate positioning processing based on HSV binary value and OTSU binary value.This method solves the problem of inaccurate positioning and unclear positioning boundaries in ordinary positioning algorithms.It provides a good foundation for license plate recognition.Aiming at the problem that the character segmentation process in the traditional license plate recognition algorithm generates a large number of errors and affects the license plate recognition rate,the convolutional neural network is applied to the character recognition of license plates.This process directly inputs the license plate as a whole into the convolutional neural network for training,and applies the trained model to identify the entire license plate.By adjusting the network parameters of CNN network depth,convolution kernel size,convolution kernel number,batch training sample number and other network parameters,the random gradient descent algorithm is used to reduce the number of batch training samples.Increasing network depth and increasing the number of convolution kernel can effectively increase the accuracy of license plate recognition to 98.6%.Aiming at the influence of various complex conditions such as strong light,weak light,distortion,rotation,motion blur and other factors on license plate recognition rate,a corresponding image enhancement method was proposed.The enhanced image is recognized by the convolution neural network.Through the comparison experiment analysis,it shows that the network structure proposed in this paper can be applied to various complex conditions and has a broad application prospect.