Anti-motion imaging photoplethysmography via self-adaptive multi-ROI tracking and selection

Area of Science:

• Biomedical Engineering
• Signal Processing
• Computer Vision

Background:

• Imaging photoplethysmography (IPPG) enables remote heart rate (HR) measurement from facial videos.
• Motion artifacts from head and facial movements significantly challenge IPPG accuracy.

Purpose of the Study:

• To develop a robust method for motion artifact reduction in IPPG.
• To enhance the accuracy and reliability of remote HR estimation.

Main Methods:

• A self-adaptive region of interest (ROI) pre-tracking and signal selection approach is proposed.
• Facial landmarks are used to define symmetric ROIs, which are tracked dynamically.
• Invisible ROIs are discarded, and signal quality is assessed via frequency spectrum features for optimal HR calculation.

Main Results:

• The proposed method demonstrates superior or comparable performance in HR estimation across multiple public datasets (PURE, VIPL-HR, UBFC-RPPG, MAHNOB-HCI).
• Root Mean Square Errors (RMSEs) were reduced, e.g., from 7.62 to 3.87 on the VIPL-HR dataset.
• The derived IPPG signals exhibit higher quality, leading to more accurate HR estimation.

Conclusions:

• The developed method significantly enhances the robustness of IPPG for real-world applications.
• This technique effectively mitigates motion disturbances, improving the practical utility of remote photoplethysmography.