Automatic Number Plate Recognition (ANPR) plays a vital role in intelligent transportation systems, but its performance for areas with bilingual license plates and harsh environmental conditions, such as Oman, has not been adequately explored. In this research, four YOLO architectures—YOLOv8n, YOLOv9t, YOLOv10n, and YOLOv11n—are compared to identify the optimum model for real-time license plate detection on the Oman ANPR dataset of 1,686 annotated images captured under various lighting and weather conditions. Each of the models was trained with consistent settings and hyperparameter-tuned configurations to be compared equitably. Accuracy, precision, recall, F1-score, and computational expense were measured thoroughly against each other. Experiments indicate that YOLOv11n performed optimally with 99.0% accuracy, 97.0% precision, and minimal computational expense (6.6 GFLOPs), confirming its suitability for edge and real-time applications. The study provides actionable insights for localized ANPR deployment and enhances Oman's efforts in developing smart cities through the validation of effectiveness of advanced YOLO models in real regional settings..
Musalhi et al. (Sun,) studied this question.