Abstract:
The paper is a comparison of the performance of selected YOLO models
(YOLOv8m, YOLOv12m, YOLOv11m, YOLOv9m, and YOLOv10m) that are
used to detect vehicle number plates in Bangladesh with respect to precision,
recall and mean Average Precision (mAP) at various Intersection over Union
(IoU) thresholds. Following the hyper-parameter tuning, YOLOv8m yields the
best performance with a higher accuracy (0.959) and recall (0.922), mAP50 of
0.946 and mAP50-95 of 0.55, which confirms its suitability to have in urban
areas. Even though YOLOv8m is better compared to others, there are other
models such as YOLOv12m, YOLOv11m, and YOLOv9m, which perform well
and can be used as well in real-time number plate recognition systems. A
comparative performance analysis presented in this paper covers a broader
spectrum of vehicle detection tasks and reveals the evidence of strengths and
weaknesses of each of the models. The results indicate that the YOLOv8m can
adapt to the traffic infrastructure and urban surveillance systems in Bangladesh
especially effectively, demonstrating the future opportunities of using deep
learning-based object detection in addressing complicated issues in urban
settings.
