Related Experiment Video
Updated: Jul 18, 2025
05:57
Blood Flow Imaging with Ultrafast Doppler
Published on: October 14, 2020
7.7K
Face2PPG: An Unsupervised Pipeline for Blood Volume Pulse Extraction From Faces
IEEE Journal of Biomedical and Health Informatics
|August 23, 2023
Summary
This study introduces robust pipelines for extracting remote photoplethysmography (rPPG) signals from facial videos. Novel methods enhance face stabilization, signal region selection, and RGB to rPPG transformation, achieving state-of-the-art results.
Area of Science:
- Biomedical Engineering
- Signal Processing
- Computer Vision
Background:
- Photoplethysmography (PPG) is crucial in medicine, well-being, and sports.
- Extracting remote PPG (rPPG) signals from facial videos offers non-contact monitoring capabilities.
Purpose of the Study:
- To develop and evaluate robust, reliable, and configurable pipelines for unsupervised rPPG signal extraction from the face.
- To introduce novel methods for improving rPPG extraction accuracy and robustness.
Main Methods:
- Evaluated state-of-the-art unsupervised rPPG processing pipelines across six datasets.
- Introduced rigid mesh normalization for face stabilization.
- Developed dynamic region selection for optimal raw signal extraction.
- Proposed Orthogonal Matrix Image Transformation (OMIT) for RGB to rPPG conversion, enhancing compression artifact robustness.
Main Results:
- All proposed methods demonstrated noticeable improvements in rPPG signal retrieval.
- Achieved state-of-the-art results for unsupervised, non-learning-based rPPG methodologies.
- Performance approached supervised, learning-based methods on some datasets.
Conclusions:
- The proposed pipeline enhancements significantly improve unsupervised facial rPPG extraction.
- Novel techniques offer robust and reliable non-contact physiological monitoring solutions.
- Findings provide valuable insights for future rPPG implementation and research.
