Autonomous face mask detection using single shot multibox detector, and ResNet-50 with identity retrieval through...

Area of Science:

• Computer Vision and Artificial Intelligence
• Public Health Technology

Background:

• The COVID-19 pandemic highlighted the need for effective public health measures, including face mask usage.
• Manual monitoring of face mask compliance is labor-intensive and impractical for large-scale enforcement.
• Computer vision offers a potential solution for automated, real-time surveillance of public health mandates.

Purpose of the Study:

• To develop an autonomous system for detecting individuals not wearing face masks in real-time.
• To implement an identity retrieval mechanism for non-compliant individuals using computer vision.
• To enhance public health enforcement through an efficient and accurate automated monitoring system.

Main Methods:

• Fine-tuning a pre-trained ResNet-50 model for binary classification of masked versus non-masked faces.
• Utilizing a Single Shot MultiBox Detector (SSD) based Caffe model for real-time face detection.
• Employing a VGG-Face based deep siamese neural network for accurate face matching and identity retrieval.

Main Results:

• The face mask detection classifier achieved a high accuracy of 99.74%.
• The identity retrieval model demonstrated strong performance with 98.24% accuracy.
• The integrated system successfully detects non-masked individuals and retrieves their identity from a database.

Conclusions:

• The proposed computer vision system offers an effective and accurate solution for real-time face mask monitoring.
• This technology can significantly reduce human effort in enforcing public health guidelines.
• The system provides a robust mechanism for identifying and managing non-compliance with face mask mandates.