Transfer learning for mobile real-time face mask detection and localization.

Area of Science:

• Computer Vision
• Public Health Technology
• Machine Learning

Background:

• The COVID-19 pandemic necessitated changes in daily habits, including mandatory face mask usage to curb transmission.
• Detecting face mask non-compliance is crucial for enforcing public health guidelines.

Purpose of the Study:

• To develop an automated method for detecting face mask usage violations in images and video streams.
• To implement a transfer learning approach for efficient and accurate face mask detection.

Main Methods:

• A transfer learning approach utilizing the MobileNetV2 model was employed.
• The method was designed to identify individuals not wearing face masks and localize the relevant facial area.
• The system was trained and evaluated on a dataset of 4095 images.

Main Results:

• The proposed method achieved a high accuracy of 0.98 in detecting face mask violations.
• The system demonstrated the ability to localize face mask detection with associated probabilities.
• Effectiveness was validated on a diverse dataset of individuals wearing and not wearing face masks.

Conclusions:

• The developed method is effective for automated face mask violation detection.
• The system operates in real-time and is suitable for deployment on devices with limited computational resources.
• This technology offers a practical solution for real-world public health monitoring.