| At present,many cities have installed integrated automatic license plate recognition systems on a large scale.Existing intelligent transportation systems achieve excellent results under certain conditions.A large number of scholars have conducted research on license plate recognition algorithms under non-specific conditions and have achieved good results.However,the accuracy of the recognition of small long-distance and inclined license plates still needs to be improved.It is of practical significance to study the above problems.The research content of this paper covers two levels.The first is to locate the license plate accurately,and the second is to recognize the characters of the license plate.The specific work content is as follows:(1)In the license plate positioning stage,in view of the problem of low detection accuracy due to the small number of long-distance license plates,an improved YOLOv4 algorithm was proposed,the network was redesigned,and the original CSPDarknet53 structure was replaced by the lightweight network Mobilenetv1 to characterize the entire image.Extraction improves the detection speed of the entire network.In addition,the channel attention mechanism is introduced to improve the accuracy of license plate positioning.The improved license plate location algorithm can effectively avoid detection errors caused by the license plate itself being far away from the camera.(2)In the license plate character recognition stage,based on the CRNN algorithm,this paper proposes an improved CRNN algorithm for the problem of low recognition rate caused by tilted license plates,and uses BGRU for prediction,so as to obtain the correlation between each data sequence,and complete the current There is a recognition task of license plate positioning,and this improvement can well avoid the impact of various deficiencies in the traditional recognition mode,fundamentally improve the current network’s recognition accuracy of license plates,and network training time and The convergence speed has been rapidly improved.In addition,a spatial attention mechanism is introduced to correct the tilted license plate image,which improves the accuracy by 2.93%. |
