A Lightweight Model for Identifying Tilted Mixed Color License Plates
The size of the license plate recognition model based on neural network is large.It is not conducive to deploy in edge computing equipment.In addition,the recognition rate is lowish for oblique license plates and mixed color license plates.To overcome the shortage,a lightweight neural network combination model is proposed for recognizing tilted mixed color license plates.Based on you only look once(YOLO)algorithm,an improved point-YOLO(P-YOLO)algorithm is presented to achieve license plate detection,classification,and localization.Then,based on license plate recognition via deep neural networks(LPRNet),an enhanced normalization-gray-LPRNet(N-G-LPRNet)algorithm is proposed to recognize license plate characters.Finally,compared to the original YOLO algorithm,the test results on the Chinese city parking dataset show that the P-YOLO algorithm significantly improves the mean average precision index for detecting tilted license plates.Combined to character recognition networks,the recognition accuracy for conventional,slightly tilted and strongly tilted license plates has improved by about 1.3％ ,70.7％ ,and 63.8％ ,respectively.In the character recognition stage,under the premise of mixed color license plate training,the N-G-LPRNet algorithm improves the recognition rates of blue and green license plates by about 40.65％ and 32.26％ ,respectively.The P-YOLO-N-G-LPRNet combination model has a comprehensive recognition rate of up to 98.16％ .Compared to YOLO-LPRNet,the performance of P-YOLO-N-G-LPRNet has increased 41.33％ .The combined algorithm for license plate recognition takes up about 5 MB,which has obvious advantages in terms of recognition rate,robustness,and lightweight,and is suitable for deployment on edge devices.
deep learningtarget detectioncharacter recognitionforecasting pointgray preprocessing
NETL
NSTL
万方数据
