Comparative Performance Analysis of YOLOv8 and YOLOv11 for Vehicle and License Plate Detection Using a Multi-Regional Dataset

Abstract

Detecting vehicles and license plates is a key part of smart transportation and monitoring systems. Detection models must operate reliably under challenging conditions such as illumination changes, occlusions, and viewpoint variations. This paper compares YOLOv8 and YOLOv11 using a multi-regional dataset from Iraq, Iran, and India. The dataset contains 16,657 training images and 1,839 validation images labeled with seven classes. Both models were trained under identical experimental settings for fair comparison. The results show that both models achieved strong detection performance, with mAP@50 values above 0.78. YOLOv8 achieved faster inference speed (47.27 FPS) and lower latency (21.16 ms/image), while YOLOv11 provided a smaller model size (18.28 MB) and fewer parameters (9.42 million). Additionally, YOLOv11 achieved slightly higher precision (0.777), whereas YOLOv8 achieved higher mAP@50 (0.791) and mAP@50–95 (0.607).