Design And Implementation Of The System For License Plate Recognition In The Haze

Hazy weather has becoming more common in several parts of China as the country’s development grows.It causes light scattering,making it impossible for various technologies to produce clear and effective images,which has a particularly negative influence in the fields of intelligent transportation,security monitoring,and other related applications.The license plate acts as the vehicle’s "portal" in the age of the Internet of Everything,and its correctness,efficacy,and speed of identification are critical in the field of transportation.This paper proposes and implements a generally applicable license plate identification system for hazy weather as a result of significant study and numerous trials,which contains the following features:(1)Haze level judgment functionFor haze-level assessment,a new neural network classification model based on deep learning is suggested.Each hyperparameter is continuously improved using haze level image dataset to improve accuracy and use the development set,test set to prove each performance of the model,to determine the best image defogging approach for various haze levels.(2)Image defogging function for different haze levelsAccording to the different levels of haze,research experiments suggest that the improved adaptive Retinex be used to remove haze for first-level haze images with severe haze,and the dark channel method with guided filtering based on different parameters be used to remove haze for second-level and third-level images with light haze.Image sharpness performance measurements and color histograms are used to highlight the method’s utility.After defogging,an automated white balance was used to restore the image’s color.Furthermore,super-resolution was used to extend the license plate region of small-size pictures for later identification.(3)license plate recognition function of multi-method fusionCombining support vector machines,Haar-like classifiers,Single Shot Multibox Detector target recognition,optimization of Open CV and library functions of Hyperlpr,as well as a large number of comparative experiments and the character classification neural network model,designing three license plate recognition modes for users to choose from: balance,fast,and accurate,demonstrating the feasibility and efficiency of the method by recognition accuracy and procedure.(4)Recognition system with good human-computer interactionPyQt was used to create the Graphical User Interface,integrate all functionalities,and create a haze-based license plate identification system that contains an image acquisition module,haze level judgment module,image defogging module,and license plate recognition module.