Scholarlink

<p>In an era of pervasive digital connectivity, protecting children's privacy in publicly shared videos has become a crucial concern. To address this problem, this paper presents an approach that leverages advanced object detection models to identify and anonymize children's faces in real-time images and videos. A custom dataset of children's faces was collected and manually labeled from publicly available websites to train and evaluate the models. In addition to detection, the approach includes a blurring mechanism to anonymize detected faces effectively, enhancing privacy protection. This paper explores the performance of three versions of the YOLO (You Only Look Once) object detection family YOLOv5, YOLOv8, and YOLOv9 by comparing their detection performance across various epochs and model sizes. Our results demonstrate the exceptional effectiveness of YOLOv9, particularly its medium-sized variant, which outperformed the other models, achieving the highest mAP@0.5 of 0.963. The findings strongly suggest that YOLOv9 is the most suitable model for real-time face anonymization, offering a promising solution to safeguard children's privacy in digital media. Furthermore, the framework aims to balance privacy concerns with usability in dynamic video environments.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhanced child face anonymization using YOLO framework for real-time privacy protection in videos

  • 期刊论文

  • 发表日期：

    2025年6月29日

查看全文
  • Hunar A. Ahmed,
  • Mohammed H. Ahmed,
  • Jafar Majidpour,
  • Saman M. Omer

摘要

In an era of pervasive digital connectivity, protecting children's privacy in publicly shared videos has become a crucial concern. To address this problem, this paper presents an approach that leverages advanced object detection models to identify and anonymize children's faces in real-time images and videos. A custom dataset of children's faces was collected and manually labeled from publicly available websites to train and evaluate the models. In addition to detection, the approach includes a blurring mechanism to anonymize detected faces effectively, enhancing privacy protection. This paper explores the performance of three versions of the YOLO (You Only Look Once) object detection family YOLOv5, YOLOv8, and YOLOv9 by comparing their detection performance across various epochs and model sizes. Our results demonstrate the exceptional effectiveness of YOLOv9, particularly its medium-sized variant, which outperformed the other models, achieving the highest mAP@0.5 of 0.963. The findings strongly suggest that YOLOv9 is the most suitable model for real-time face anonymization, offering a promising solution to safeguard children's privacy in digital media. Furthermore, the framework aims to balance privacy concerns with usability in dynamic video environments.