RESEARCH YOLOv8 AND YOLO-NAS VERSIONS IN LICENSE PLATE DETECTION

Abstract

Recently, license plate recognition systems have been an important part of many traffic management and security systems such as automatic speed control, stolen vehicle tracking, automatic toll management, and control of vehicles entering and exiting bus station areas, schools, hospitals, etc. During the research process, we compared versions of YOLOv8 and YOLO-NAS based on the criteria of Accuracy, Precision, Recall, and F1 score to evaluate the most suitable models for vehicle license plate recognition in Vietnam under different environmental conditions. This review provides perspective for developers or last users to choose the most suitable technique for their application. The results show that for applications with good infrastructure and high accuracy requirements, YOLO-NAS-S is a suitable model with an Accuracy of 83.92%, Precision of 0.9125; Recall is 0.9125, and F1 score is 0.9125. For less developed infrastructure and speed requirements, YOLOv8n can be used with a smaller number of parameters but the accuracy is acceptable, Accuracy is 81.4%; Precision is 0.9625; Recall is 0.8415, and F1 score is 0.8979.