Research On License Plate Detection And Recognition Algorithm In Haze Weather

With the rapid development of China's economy,the haze weather caused by industrial and vehicle exhaust gas appears frequently,atmospheric turbidity and low visibility cause image detection and acquisition inaccuracy,which has a serious impact on the intelligent transportation system and licence plate recognition.This subject conducts research on the license plate detection and recognition algorithm in fog and haze,and aims to solve the problems of low recognition rate,slow detection speed,and poor robustness of the license plate detection and recognition system in fog and haze,so as to improve traffic control ability to ensure the safety and convenience of people traveling in hazy days.According to the algorithm steps of image defogging,license plate positioning,license plate character segmentation and recognition,respectively,this paper conducts deep research to realize the license plate detection and recognition algorithm in haze weather.Image defogging:this part studied the dark channel prior defogging algorithm and the transmission image optimization algorithm based on guided filtering,and an improved dark channel prior defogging algorithm is proposed to solve the shortcomings of brightness slants dark and poor timeliness.Firstly,the coarse transmittance is optimized by using the fast guiding filtering algorithm,and the atmospheric light intensity is also refined according to the proportion of airspace in the image.Then the two-dimensional gamma adaptive algorithm is used to equalize the brightness of the image after the fog removal,so as to obtain a clear fog removal image.License plate positioning:for the image after fog removal,the algorithm of mathematical morphology operation and edge detection is combined to coarsely locate the license plate,and then the precise position of the license plate is selected according to the connected area analysis method;license plate character segmentation and recognition:the horizontally corrected tilted license plate is corrected by the Hough transform algorithm,And select the projection-based license plate character segmentation algorithm to improve the accuracy and effectiveness of the license plate character segmentation,and then studied the structure of classic digital identification LeNet-5 convolution neural network,based on the characteristics of license plate character in china,it also adjust the structure of the network,select the better ReLU activation function,add dropout strategy in the hidden layer,put forward the improved LeNet-5 convolution neural network recognition algorithm,in the end,realize the precise identification of license plate characters in haze weather.The experimental comparison shows that the dark channel prior defogging algorithm based on fast guided filter has a great improvement in defogging effect and timeliness,the optimized LeNet-5 network structure has the characteristics of high localization and good recognition rate.By using in license plate image detection of real foggy environment,the results show that the algorithm has achieved remarkable results in recognition rate,timeliness and robustness.