Developing License Plate Recognition System Using Advanced SCRFD and LPRNet

The paper focuses on solving three main problems in license plate recognition in intelligent traffic. The first is to detect and classify vehicles right at the red light waiting area. The second is to detect license plates from cars detected from problem one. Finally, detect the character on the license plate. We have selected the SCRFD model - a well-known model for face detection for the first two problems of vehicle detection and license plate detection. This paper modified this vehicle and license plate detection model based on key points. The model is trained with vehicle image data on Vietnamese streets to improve the performance of vehicle and license plate detection. The results show that the SCRFD model has an accuracy of 99.48% with a processing speed of 17.075 ms, which is suitable for real-time requirements and achieves good results even in the case of heavy traffic. We use the LPRNet model to simulate character recognition for the third problem of character detection. As a result, the system could recognize with 95% accuracy and a processing speed of 3 ms. This shows that the system is feasible when applied in practice.