HRMamba: Fusing Luminance Information for Remote Physiological Measurement in Varied Lighting Conditions.

Area of Science:

• Biomedical Engineering
• Computer Vision
• Physiological Monitoring

Background:

• Camera-based photoplethysmography (cbPPG) offers non-invasive vital sign monitoring via facial videos.
• Deep learning methods struggle with weak signal extraction, long-range dependencies, and varying illumination in cbPPG.
• Challenges include spatial-temporal redundancy and accurate PPG extraction in complex lighting, especially low-light scenarios.

Purpose of the Study:

• To develop an end-to-end deep learning method for robust cbPPG signal extraction.
• To address challenges in weak PPG signal detection and periodic pattern recognition.
• To improve PPG signal accuracy under diverse and challenging lighting conditions.

Main Methods:

• Proposed HRMamba, an end-to-end method based on Mamba architecture.
• Employed temporal difference Mamba for temporal signal processing and bidirectional state space for robust pattern learning.
• Introduced a luminance post-processing module to extract and embed luminance information without altering original video data.

Main Results:

• HRMamba achieved state-of-the-art performance in cbPPG signal extraction.
• The luminance post-processing module demonstrated effectiveness across various lighting conditions.
• Performance significantly improved in dark environments without compromising normal light scene accuracy.

Conclusions:

• HRMamba provides a robust and effective solution for cbPPG vital sign monitoring.
• The proposed method overcomes key limitations of existing deep learning approaches for cbPPG.
• The luminance module offers a versatile enhancement for low-light cbPPG applications.