A Comparative Survey of Methods for Remote Heart Rate Detection From Frontal Face Videos

Area of Science:

• Physiological monitoring
• Affective computing
• Biomedical signal processing

Background:

• Remote physiological measurement offers significant advantages in medical and affective computing.
• Current methods for unobtrusive heart rate (HR) detection from facial recordings utilize subtle color changes or motions.
• Existing approaches lack consensus due to varied comparison datasets.

Purpose of the Study:

• To systematically review and evaluate remote HR detection methods from facial recordings.
• To analyze algorithms across the HR processing pipeline: video processing, blood volume pulse (BVP) signal extraction, and HR computation.
• To compare method performance using the MAHNOB-HCI public database.

Main Methods:

• Literature review of HR detection methods from 2008 to present.
• Classification of approaches into face video processing, BVP signal extraction, and HR computation stages.
• Performance analysis of algorithms using the MAHNOB-HCI dataset.

Main Results:

• Extracted facial skin areas demonstrate richer blood volume pulse (BVP) information.
• Blind source separation and peak detection techniques exhibit robustness against head motion.
• Comparative analysis identified strengths and weaknesses of various algorithms within the HR pipeline.

Conclusions:

• Remote HR monitoring via facial analysis is feasible and beneficial.
• Specific signal processing techniques, particularly blind source separation and peak detection, are effective for HR estimation.
• Further research should focus on robust methods for real-world applications with motion artifacts.