High-Fidelity rPPG Waveform Reconstruction from Palm Videos Using GANs.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Computer Vision

Background:

• Remote photoplethysmography (rPPG) facilitates non-contact physiological monitoring using cameras.
• Existing rPPG research often prioritizes metric estimation over waveform reconstruction.
• Current datasets with fingertip PPG labels limit facial rPPG waveform accuracy.

Purpose of the Study:

• To develop a robust method for reconstructing high-fidelity rPPG waveforms.
• To address limitations of existing datasets and models in rPPG waveform generation.
• To establish a reliable foundation for advanced physiological signal analysis using rPPG.

Main Methods:

• Collected a novel dataset pairing palm-region videos with wrist-based PPG signals.
• Proposed a generative adversarial network (GAN)-based pulse-wave synthesis framework.
• Incorporated time-domain peak-aware loss, frequency-domain loss, and adversarial loss for denoising.

Main Results:

• Achieved promising performance with RMSE of 0.102 and MAPE of 0.028.
• Demonstrated high signal fidelity with Pearson correlation of 0.987 and cosine similarity of 0.989.
• Validated the effectiveness of the new dataset and GAN framework for rPPG waveform reconstruction.

Conclusions:

• The proposed approach successfully reconstructs high-fidelity rPPG waveforms with improved morphological accuracy.
• The novel dataset and GAN framework offer a reliable foundation for rPPG-based health applications.
• This work advances rPPG waveform reconstruction, enabling more accurate physiological signal analysis.