Abstract:
This work presented here compares the performance of various YOLO models,
including YOLOv8m, YOLOv12m, YOLOv11m, YOLOv9m, and YOLOv10m. The
models are compared in the context of Bangladeshi vehicle number plate
detection based on the leading metrics of precision, recall, and mean Average
Precision (mAP) at various Intersection over Union (IoU) thresholds. The study
conducts using hyper-parameter tuning YOLO models. These experiments show
that YOLOv8m is the best with the maximum precision (0.959) and recall (0.922)
with mAP@50 as 0.946, and mAP@50-95 as 0.55. This shows the effectiveness of
YOLOv8m in detecting vehicle number plates in varied urban environments.
Though YOLOv8m is better performing, YOLOv12m and other models like
YOLOv11m and YOLOv9m also perform competitively and hence can be used in
real-time vehicle number plate detection systems. YOLOv8m's better overall
performance highlights its viability in traffic management system and
surveillance applications, making it a desirable choice for Bangladeshi vehicle
number plate detection software. This work enlightens the application of YOLO
models in using them for the task of vehicle detection and extends the limits in
urban object detection through deep learning.
