Research On License Plate Recognition Technology Based On Improved Faster R-CNN

With the continuous advancement of technology,the field of intelligent transportation is also constantly innovating.The rapid development of today’s intelligent transportation field is inseparable from driving license detection and recognition algorithms.In recent years,research in this area has made breakthroughs.However,although the existing algorithms have good recognition effects in simple scenes,the recognition effects of the existing algorithms are still not ideal in complex and changeable scenes such as night,ice,snow,tilt,and blur.Therefore,it is still a challenging task to accurately detect and recognize license plates.The main difficulty lies in the text interference in natural scenes(such as billboards,road signs,etc.)and complex and changeable environments such as different lighting and tilt,ice and snow occlusion and blurring,etc.To this end,this paper has carried out research on license plate recognition technology based on improved Faster R-CNN(Faster Regions with CNN)for license plate recognition in different environments.First,the thesis analyzes the current research status of traditional license plate recognition algorithms at home and abroad.The algorithm steps are roughly divided into license plate location,character segmentation,and character recognition.Because each step is related to each other,it is easy to enlarge and superimpose small errors,resulting in a greatly reduced accuracy.Therefore,in order to improve the accuracy of the license plate recognition algorithm,the character segmentation stage should be skipped to avoid error amplification.Secondly,the accuracy of license plate location directly affects the recognition rate of the entire license plate recognition algorithm,so this paper uses Res Net network as the backbone network to reduce information loss and loss,thereby protecting the integrity of information.Use RNN+BLSTM+CTC to enhance the accuracy of character recognition.The data set uses the open source Chinese License Plate Data Set(CCPD),from which more than 80,000 license plate photos classified by night,inclination,blur,snow,etc.,were selected,and then four were produced through the assistance of Model Arts and manual labeling methods.Data sets for different environments.Finally,the network model proposed in this paper is trained and tested,and compared and analyzed with other license plate recognition algorithms.It can be seen from the experimental results that in a complex and changeable environment,the end-to-end license plate recognition network proposed in this paper has a higher recognition rate and accuracy.