A Comparative Study of Principled rPPG-Based Pulse Rate Tracking Algorithms for Fitness Activities

Area of Science:

• Physiological monitoring
• Biomedical signal processing
• Computer vision

Background:

• Remote photoplethysmography (rPPG) advancements for pulse rate (PR) estimation often lack module-level analysis.
• Existing evaluations primarily focus on system-level performance, hindering understanding of individual module contributions.
• Explainable AI and physiological monitoring research requires detailed insights into rPPG algorithm components.

Purpose of the Study:

• To conduct a comparative study of principled video-based PR tracking algorithms.
• To analyze module contributions within a reconfigurable rPPG sensing system.
• To evaluate performance in challenging fitness scenarios with varying signal-to-noise ratios (SNR).

Main Methods:

• Constructed a reconfigurable rPPG sensing system based on a decade of research.
• Performed experiments on two datasets: internal (25 videos) and public ECG Fitness (34 videos) with diverse exercise types.
• Evaluated performance using signal-to-noise ratio (SNR), Pearson's correlation, error count ratio, error rate, and root mean squared error (RMSE).

Main Results:

• The top-performing configuration achieved 0.8 dB SNR, 0.86 correlation, 9% error rate, and 3.3 bpm RMSE on the internal dataset.
• On the ECG Fitness dataset, the best configuration yielded 1.3 dB SNR, 0.77 correlation, 6.0% error rate, and 8.1 bpm RMSE.
• Demonstrated significant improvements over alternative configurations, indicating synergistic effects.

Conclusions:

• A synergistic effect was observed between pulse color mapping and adaptive motion filtering for PR estimation.
• Robust frequency tracking algorithms are crucial for accurate PR estimation in low SNR conditions.
• The study provides a foundation for explainable AI in physiological monitoring through detailed module analysis.