Robust real-time heart rate prediction for multiple subjects from facial video using compressive tracking and support...

Area of Science:

• Biomedical Engineering
• Physiological Monitoring
• Computer Vision

Background:

• Remote monitoring of vital signs offers nonintrusive health assessment.
• Image photoplethysmography (iPPG) enables contactless physiological measurement.
• Accurate heart rate (HR) estimation is crucial for health surveillance.

Purpose of the Study:

• To develop and validate a robust, self-learning iPPG-based method for remote HR estimation.
• To improve the accuracy and reliability of noncontact HR monitoring.
• To assess the performance of the method across multiple subjects and simultaneous monitoring scenarios.

Main Methods:

• Utilized an improved compress tracking algorithm for region of interest tracking in video sequences.
• Employed a support vector machine (SVM) classifier to filter false heartbeats.
• Implemented a self-learning procedure for adaptive iPPG signal processing.
• Conducted experiments with 40 subjects for individual HR estimation and 10 subjects for simultaneous monitoring.

Main Results:

• Reduced the mean absolute error in HR estimation from 3.6 to $$.
• Achieved real-time HR prediction from video sequences at 600x800 resolution.
• Demonstrated modest but significant effects on HR prediction accuracy.
• Validated the self-learning approach for enhanced iPPG signal processing.

Conclusions:

• The developed iPPG method provides a robust and accurate approach for remote, noncontact HR monitoring.
• The self-learning procedure and SVM filtering significantly improve the reliability of iPPG-based HR estimation.
• The system demonstrates potential for real-time, multi-subject physiological monitoring applications.