ResoPhys: Unsupervised Plug-and-Play Remote Physiological Measurement via Facial Videos of Arbitrary Resolution.

Area of Science:

• Biomedical Engineering
• Computer Vision
• Signal Processing

Background:

• Remote photoplethysmography (rPPG) offers non-contact vital sign monitoring, but is limited by fixed-resolution input requirements.
• Real-world applications face challenges with arbitrary-resolution videos due to varying camera distances.

Purpose of the Study:

• To introduce ResoPhys, an unsupervised, plug-and-play method for rPPG analysis in arbitrary-resolution facial videos.
• To enable robust rPPG measurements irrespective of video resolution.

Main Methods:

• ResoPhys employs random video scaling to create multi-resolution pairs.
• Specialized modules handle arbitrary-resolution feature extraction and upsampling.
• Unsupervised contrastive learning with a novel multi-resolution loss optimizes the framework.

Main Results:

• ResoPhys demonstrates superior performance compared to existing unsupervised methods across various resolutions.
• The method shows particular strength in low-resolution scenarios, proving robustness to resolution changes.
• Evaluated on public datasets, ResoPhys validates its effectiveness.

Conclusions:

• ResoPhys overcomes the resolution limitation in rPPG, making it suitable for practical telemedicine and health monitoring.
• The plug-and-play nature of ResoPhys allows seamless integration with existing rPPG networks.
• This work provides a universal front-end for rPPG systems adaptable to diverse video resolutions.