Thai License Plate Recognition Based on YOLO Detector and CNN-Classifier

Abstract

License Plate Recognition (LPR) has many different applications, especially in the intelligent transportation system, for example, an automatic tollgate that allows people to pass through and then bills the cost later according to the identification captured by the LPR system. Although a significant amount of research has been done on Western and Chinese license plates, there is a lack of work on Thai license plates. In this work, we introduce an LPR system that focuses on Thai license plates using a combination of YOLOv5 (You Look Only Once) object detection, mathematical morphology, and Convolutional Neural Network image classification. The proposed method is trained using images of vehicles that are presented at tollgate. The plate detection model is trained with 400 images while the OCR model is trained using 23003 images. In testing, the system achieves 98.4% recall for plate detection, 98.22% accuracy for character recognition, and 93.37% accuracy in the end-to-end scenario.