ConDiff-rPPG: Robust Remote Physiological Measurement to Heterogeneous Occlusions

Area of Science:

• Biomedical Engineering
• Computer Vision
• Signal Processing

Background:

• Remote photoplethysmography (rPPG) enables contactless physiological and cardiac monitoring via facial videos.
• Real-world occlusions (e.g., movement, dust) degrade the accuracy of existing rPPG methods.
• Robust rPPG measurement is crucial for widespread clinical and consumer applications.

Purpose of the Study:

• To develop a novel method, ConDiff-rPPG, that improves the robustness of rPPG measurements against various occlusions.
• To enhance the accuracy and reliability of contactless physiological monitoring in unconstrained environments.

Main Methods:

• A diffusion model (ConDiff-rPPG) was designed to recover missing information in occluded facial videos.
• Spatio-temporal video representation with masks was used to compress damaged face data.
• Low-rank decomposition regularization was incorporated to reduce background noise and enhance informative features.

Main Results:

• ConDiff-rPPG demonstrated superior robustness and accuracy in rPPG measurements compared to existing methods when facing occlusions.
• Extensive experiments, including cross-dataset evaluations, validated the model's generalization ability.
• Ablation tests confirmed the effectiveness of the proposed components in improving rPPG performance.

Conclusions:

• ConDiff-rPPG significantly enhances the reliability of remote photoplethysmography in the presence of common real-world occlusions.
• The proposed method offers a promising solution for accurate contactless physiological monitoring in diverse scenarios.
• This work advances the field of rPPG by addressing a critical limitation and improving practical applicability.