Complicated Condition License Plate Recognition Based On Convolutional Neural Network

Automatic license plate detection and recognition technology plays an important role in intelligent transportation.The use of automatic license plate recognition technology can realize intelligent management of vehicles in various scenarios such as urban communities and parking lots.In recent years,due to the huge potential applications of automatic license plate detection and recognition technology,researchers have attracted extensive attention and exploration.The current license plate recognition technology is mostly carried out under controlled conditions.For example,the common urban district license plate recognition is to obtain better pictures under a specific location and light.It has achieved better recognition results.The license plate recognition technology on the road often It is through the bright light to obtain better visual conditions for positioning recognition.Due to the harsh recognition conditions,the current license plate recognition technology often cannot achieve good recognition results in complex environments.Therefore,it is still a difficult challenge to recognize license plates in complex environments.The difficulty of license plate recognition in a complex environment is mainly to recognize license plates that are far away,or to recognize license plates that rotate at a large angle.The early license plate recognition process is divided into two parts.First,the license plate is detected and segmented,and then the characters on the segmented picture are recognized one by one.Since the detection of the license plate and the recognition of the license plate have a considerable degree of correlation,obtaining a more accurate positioning in the license plate detection stage can improve the accuracy of the character recognition of the license plate,and vice versa.This article mainly conducts related research on the location and recognition of the license plate under the condition of long shooting distance and rotation.In order to facilitate the description,the above two situations are defined as complex conditions.This article mainly focuses on complex conditions.Car license plate recognition research,the specific research content is as follows:(1)This thesis studies the license plate location algorithm based on Faster R-CNN.In this dissertation,Faster R-CNN algorithm is used to locate the license plate under complex environment.To analyze the situation where the license plate with a long shooting distance cannot be detected,in order to enable the smaller target features to be retained in the final feature data,this thesis uses the method of feature fusion on different convolutional layers to adjust the proportion and size of the anchor to increase Recall of small goals.The mapping function of the RPN layer was redesigned to realize the use of three parameterized vertex coordinate regression to surround the parallelogram of the license plate.The improved license plate positioning algorithm not only has a better recognition effect for the license plate with a longer shooting distance.The use of parallelograms to enclose the license plate is beneficial to correct the rotating license plate.(2)Realize the recognition of license plate characters.This thesis recognizes the license plate characters based on CRNN algorithm.To analyze the problem of low accuracy of large-angle license plate recognition,this paper corrects the large-angle license plate and divides the license plate into Chinese characters and non-Chinese characters,and uses the improved LeNet-5 network for the Chinese license plate.Recognize,for the non-Chinese part,use CRNN network for recognition.Through experiments,we found that the recognition of the license plate after segmentation has a greater improvement in accuracy than the direct use of the CRNN network recognition.