A Statistical Evaluation and Application of YOLO-Based Models for Cross-Language License Plate Recognition: Arabic and Latin Alphabets

Abstract

This study presents a statistical evaluation and application of YOLO-based models for cross-language license plate recognition. License plate recognition is essential for enforcing traffic regulations and reducing traffic accidents. A dataset comprising 1,834 images was utilized to evaluate and assess the performance of the proposed techniques. Despite significant advancements in machine learning, this task remains challenging, particularly in regions where license plates feature diverse languages and scripts, such as Arabic and Latin alphabets. We compared two models, YOLOv5 and YOLOv8, and found that YOLOv8 outperforms YOLOv5, achieving an accuracy of 96.1% compared to YOLOv5’s 94.1%. This paper highlights the efficacy of the proposed license plate recognition system in addressing the challenges associated with Arabic and Latin license plates, demonstrating remarkable performance in detecting, segmenting, and recognizing license plate numbers.  This research emphasizes both the practical application and quantitative evaluation of YOLO-based models, demonstrating that the proposed system provides reliable and high-performing solutions for license plate recognition across different languages and scripts. These findings underscore the potential of combining deep learning with rigorous statistical analysis for applied computer vision tasks.