Automatic detection of face mask wearing based on polarization imaging

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• The global rise in respiratory illnesses necessitates effective face mask detection.
• Developing automatic systems for face mask detection presents significant challenges.
• Polarization imaging offers unique properties for enhanced object detection.

Purpose of the Study:

• To propose a novel face mask detection framework, DOLP-YOLOv5.
• To leverage polarization imaging for improved face mask detection accuracy.
• To enhance feature extraction and fusion for robust detection.

Main Methods:

• Developed DOLP-YOLOv5 framework integrating polarization imaging.
• Incorporated lightweight shuffle attention (SA) for feature extraction.
• Utilized Content-Aware Bidirectional Feature Pyramid Network (CA-BiFPN) for feature fusion.
• Employed Focal-EIoU loss for precise bounding box regression.

Main Results:

• DOLP-YOLOv5 achieved a mAP50-95 of 63.5% on the PFM dataset.
• Demonstrated performance improvements over YOLOv8s (3.08%) and YOLOv9s (4.44%).
• Achieved a high response speed of 384.6 frames per second.

Conclusions:

• DOLP-YOLOv5 shows superior performance in face mask detection.
• Polarization imaging is effective for enhancing face mask detection.
• The framework has potential applications in other polarization imaging detection tasks.