Research And Implementation Of License Plate Recognition Method Under Fog Constraint

Intelligent transportation system is an effective way of traffic management,it realizes the way of real-time monitoring traffic conditions,not only improve the quality of traffic management and reduce the number of staff.License plate recognition as an important part of intelligent transportation system has also been more and more attention.After 20 years of development,license plate recognition has been widely used in traffic intelligent monitoring,residential access control,non-stop automatic charges and other occasions.But in the case of fog,the license plate recognition by the interference of atmospheric light and the rate of decline,which makes intelligent traffic can’t work.Image dehazing algorithm is a way to remove fog interference and restore clear images.In this paper,the image dehazing is used to eliminate the interference of the fog and then the license plate recognition is carried out.The main contents are as follows.In the stage of image haze removal,this paper first introduces the image degradation model,and then describes two degenerative models based on the image haze removal method,dark channel a priori method and non-local image dehazing method.In this paper,the bright region is judged by the absolute value of the atmospheric light and the dark channel value,and the transmittance of the bright region is modified by the non-local image haze removal algorithm when the image is restored in some bright areas.Through a large number of experiments,it is found that the adjustment factor in the transmittance smoothing formula has some influence on the image restoration result,so this paper uses an adaptive adjustment factor to smooth the transmittance graph.In the license plate positioning stage,this paper combine the license plate color feature and the edge feature to extract the license plate candidate area,and adjust the size of all the license plates,and then and then the initial screening to the candidate license plate according to the area,length-width ratio and other features.The horizontal and vertical skew corrections are performed on the selected license plates.Finally,the trained SVM classifiers are used to classify the corrected license plates to obtain the final license plate.In the license plate recognition stage,this paper first combine the connected area extraction method and the vertical projection method to divide the license plate image into seven single characters,and adjust the character size to 24 * 24,and then through the improved LeNet-5 convolution neural network model to identify.The improved convolution neural network model has millions of parameters,in order to speed up the convergence rate of the model,this paper adds the data normalization layer in the first convolution layer and the second convolution layer respectively.However,the convergence of the model is also affected by the learning rate update strategy.Therefore,this paper analyzes the relationship between the change trend of the learning rate and the number of iterations in detail.Finally,the inv method is used to update the learning rate,and the best recognition rate on the verification set when the model converges is 98.7%In this paper,the recognition rate of the license plate recognition algorithm under the fog constraint is 88.61%.