Combining Video Magnification with Machine Learning-Based Source Identification for Contactless Heart Rate Monitoring

Area of Science:

• Biomedical Engineering
• Computer Science
• Signal Processing

Background:

• Conventional heart rate (HR) monitoring methods face challenges including patient discomfort, skin irritation, and poor adherence.
• Contactless sensing systems are needed to overcome limitations of traditional HR monitoring.

Purpose of the Study:

• To develop and validate a robust hybrid framework for accurate contactless heart rate estimation from facial video.
• To enhance HR estimation accuracy by combining advanced signal processing and machine learning techniques.

Main Methods:

• A two-stage geometric stabilization pipeline with dense facial tessellation was used to mitigate motion artifacts.
• Eulerian Video Magnification (EVM) and chrominance-based Region of Interest (ROI) filtering were employed for signal enhancement.
• Signal recovery involved sliding-window Principal Component Analysis (PCA) and Second-Order Blind Identification (SOBI), with a Light Gradient Boosting Machine (LightGBM) classifier for physiological source identification.

Main Results:

• The framework achieved a Mean Absolute Error (MAE) of 1.50 bpm, Root Mean Square Error (RMSE) of 3.07 bpm, and a Pearson Correlation Coefficient (PCC) of 0.97 on the COHFACE dataset.
• The method demonstrated robustness across diverse lighting conditions.
• Performance was comparable to state-of-the-art deep learning models and superior to traditional algorithms.

Conclusions:

• The proposed hybrid framework offers an accurate and robust solution for contactless heart rate monitoring using facial video.
• This interpretable approach provides a viable alternative for long-term, non-invasive health monitoring.
• The study highlights the potential of advanced signal processing and machine learning in remote physiological sensing.