BRPDNet: A BioRegion Prompt Distillation Network for Physiological Monitoring.

Area of Science:

• Computer Vision
• Biomedical Signal Processing
• Machine Learning

Background:

• Extracting physiological signals from video is difficult in dynamic and occluded settings.
• Current methods often fail to balance accuracy with model efficiency, especially with facial occlusions or signal redundancy.

Purpose of the Study:

• To develop an efficient framework (BRPDNet) for physiological signal extraction from video.
• To improve accuracy, robustness, and real-time performance in challenging environments.

Main Methods:

• Proposed BRPDNet framework with BioRegion Prompt module for adaptive convolution.
• Incorporated Hyper Distillation module to reduce signal redundancy.
• Utilized a teacher-student network structure for occlusion adaptability and reduced complexity.

Main Results:

• BRPDNet demonstrated superior accuracy, robustness, and efficiency compared to state-of-the-art models.
• Achieved a Mean Absolute Error (MAE) of 1.55 bpm and a Pearson Correlation Coefficient (PCC) of 0.76 on PURE and UBFC-rPPG datasets.
• Outperformed existing models with fewer parameters, ensuring efficient real-time processing.

Conclusions:

• BRPDNet offers a robust and efficient solution for physiological signal extraction from video.
• The framework effectively handles dynamic and occluded environments without explicit segmentation.
• Achieved high performance metrics with reduced computational complexity.