Research On Key Technologies Of Blurred License Plate Image Recognition In Surveillance Videos

As one of the most important parts of intelligent transportation system,license plate recognition plays an irreplaceable role in people's daily life.However,due to the defects of surveillance cameras,bad weather,the relative motion between vehicles and cameras and so on,some of captured license plate images usually get blurred,thus,it is necessary to establish the application system which can recognize blurred license plate images.In this paper,based on the detailed analysis of blurred license plate images' specific features and the deep research of license plate recognition,image blur assessment,blurred image recognition,a recognition system for blurred plate images is established.The main work and research findings of this paper can be concluded as the following three aspects:Firstly,the common methods of license plate location,character segmentation and preprocessing are summarized,and combined with specific features of these blurred license plate images to be recognized,the corresponding method of location,character segmentation and preprocessing for blurred license plate images are proposed,then based on these methods a blurred license plate character image database is constructed.Secondly,the blur degree of blurred license plate images is assessed in this paper.The purpose of blur assessment is to distinguish severely blurred license plate images from slightly blurred license plate images,so as to design recognition system only for slightly blurred license plate images.Firstly,a subjective assessment model is established based on the recognition ability of human eye to blurred plate images' rear five characters,and a blurred license plate image database is constructed on the basis of the established subjective model;secondly,based on the constructed database,an objective assessment model is established by using projective dictionary pair learning model and logistic regression model.The experimental results show that the assessment results of the objective assessment model has a good consistency with the subjective model.Thirdly,based on the close combination of sparse representation based classification model and blind deblurring model,a blurred license plate character recognition model is established.As image deblurring and image recognition are coupled in an energy equation,the established model makes image recognition task and image deblurring task benefit from each other,and achieves simultaneous restoration and recognition of blurred character images.The experimental results show that the proposed model has good deblurring and recognition results.