Scholarlink

In response to the growing need for public health measures, face masks have become a widely adopted method for reducing the transmission of respiratory illnesses. Recognizing the importance of monitoring mask adherence, this study employs computer vision to create an effective surveillance system. The proposed model relies on a sophisticated approach, utilizing a hyper-tuned deep neural network (DNN) to identify human faces with and without masks. Initially, the DNN extracts the region of interest corresponding to the face. After this, transfer learning is used with hyper-tuned MobileNetV2 classifier. The proposed model is trained with datasets, containing 11,042 and 45,000 images of two categories, viz., face with and without face mask. The proposed technique achieved 99% and 97% accuracy with and without the mask, respectively. Results were compared with existing models and validated with images of 112 live students, in both cases the proposed technique has shown significant improvement.

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

Hyper-Tuned Deep Neural Network-Based Surveillance Method for Human Face with and Without Face Masks

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Aquib Azhar,
  • Jasvinder Singh Bhatti,
  • Yash Paul,
  • Gurpreet Kaur,
  • Satwinder Singh

摘要

In response to the growing need for public health measures, face masks have become a widely adopted method for reducing the transmission of respiratory illnesses. Recognizing the importance of monitoring mask adherence, this study employs computer vision to create an effective surveillance system. The proposed model relies on a sophisticated approach, utilizing a hyper-tuned deep neural network (DNN) to identify human faces with and without masks. Initially, the DNN extracts the region of interest corresponding to the face. After this, transfer learning is used with hyper-tuned MobileNetV2 classifier. The proposed model is trained with datasets, containing 11,042 and 45,000 images of two categories, viz., face with and without face mask. The proposed technique achieved 99% and 97% accuracy with and without the mask, respectively. Results were compared with existing models and validated with images of 112 live students, in both cases the proposed technique has shown significant improvement.