DEVELOPING A SYSTEM FOR DETECTING AND RECORDING LICENSE PLATES OF MOTORCYCLISTS VIOLATING HELMET REGULATIONS

Abstract

This thesis presents a comprehensive solution for detecting helmet violations among motorcyclists and extracting their license plate information for traffic monitoring applications. The methodology centers on the integration of YOLO11 for detecting motorcyclists, heads, and license plates. To further enhance accuracy, the system incorporates preprocessing steps like perspective transformation to correct skewed license plates, ensuring reliable text recognition even under challenging conditions. While PaddleOCR is utilized for comparative analysis, the system's primary focus is on leveraging YOLO11 for superior performance in detection and recognition tasks. The proposed system was trained and evaluated on a diverse dataset, achieving high precision and recall, effectively detecting helmet violations and accurately extracting license plate numbers. The results highlight YOLO11's capability to handle small, lowresolution, or obliquely angled plates, outperforming PaddleOCR in complex scenarios. This study confirms the viability of using advanced object detection models for realworld traffic enforcement systems and concludes with recommendations for future enhancements, including integrating database storage, expanding datasets, and optimizing real-time deployment.