MDAR: A Multiscale Features-Based Network for Remotely Measuring Human Heart Rate Utilizing Dual-Branch Architecture...

Area of Science:

• Biomedical Engineering
• Computer Vision
• Signal Processing

Background:

• Remote photoplethysmography (rPPG) offers non-contact heart rate monitoring using facial video analysis.
• Challenges include ambient light variations, facial movements, and differing light absorption/reflection.
• Existing deep learning methods struggle with these real-world rPPG complexities.

Purpose of the Study:

• To develop a robust and efficient deep learning model for accurate remote heart rate measurement.
• To address the limitations of current rPPG techniques in variable conditions.
• To improve the reliability of contactless heart rate monitoring.

Main Methods:

• Proposed a multiscale feature-based heart rate measurement network (MDAR).
• Implemented a dual-branch framework combining static and dynamic facial features.
• Introduced an alternate time-shift module for enhanced temporal depth and multiscale feature fusion.

Main Results:

• MDAR demonstrated fast inference speed and significantly improved performance on UBFC-rPPG, PURE, and MMPD datasets.
• Achieved at least 30.6% improvement in Mean Absolute Error (MAE) and 30.2% in Mean Absolute Percentage Error (MAPE).
• Outperformed existing state-of-the-art methods in accuracy and reliability.

Conclusions:

• MDAR offers a robust and efficient solution for remote photoplethysmography.
• The proposed methods enhance signal processing for contactless heart rate monitoring.
• MDAR shows considerable potential for practical applications in health management and activity monitoring.