HBP-net for robust remote heart rate estimation using heartbeat probability.

Area of Science:

• Biomedical Engineering
• Computer Vision
• Physiological Monitoring

Background:

• Remote photoplethysmography (rPPG) offers contactless heart rate monitoring.
• Existing rPPG methods are sensitive to motion and lighting inconsistencies.
• Current approaches often require complex blood volume pulse signal reconstruction.

Purpose of the Study:

• To develop a more robust rPPG method resilient to motion and lighting changes.
• To simplify the heart rate estimation process by focusing on heartbeat detection.
• To improve the feasibility of camera-based vital sign monitoring in real-world applications.

Main Methods:

• Developed HBP-Net, a novel approach reframing rPPG as a heartbeat detection problem.
• Utilized a spatiotemporal attention architecture for direct heartbeat probability prediction from facial video.
• Evaluated performance on multiple datasets, including a new benchmark designed for motion challenges.

Main Results:

• HBP-Net achieved competitive accuracy under static conditions.
• The method demonstrated maintained performance with increasing motion.
• The approach bypasses the need for full blood volume pulse signal reconstruction, reducing computational complexity.

Conclusions:

• HBP-Net offers a conceptually simpler and more resilient framework for rPPG.
• The probabilistic event detection approach enhances robustness against real-world environmental variations.
• This advancement supports reliable camera-based vital sign monitoring in telehealth, fitness, and patient assessment.