A Motion and Illumination Resistant Non-Contact Method Using Undercomplete Independent Component Analysis and...

Area of Science:

• Biomedical Engineering
• Signal Processing
• Computer Vision

Background:

• Conventional heart rate (HR) estimation requires skin contact, limiting applications.
• Remote photoplethysmography (rPPG) offers non-contact HR monitoring using facial videos.
• Existing rPPG methods struggle with motion and illumination artifacts.

Purpose of the Study:

• To develop a robust rPPG method for accurate heart rate estimation.
• To address limitations of existing methods in handling motion and illumination variations.
• To propose a novel approach for Blood Volume Pulse (BVP) signal extraction.

Main Methods:

• Utilized undercomplete independent component analysis for BVP extraction.
• Employed a non-linear Cumulative Density Function (CDF) optimized via a customized Levenberg-Marquardt algorithm (U-LMA).
• Tested the U-LMA method under constrained, motion, and illumination variation scenarios.

Main Results:

• The U-LMA method demonstrated high performance with significant Pearson correlation coefficients.
• Bland-Altman plots showed small statistical limits, indicating excellent agreement.
• Comparative analysis confirmed U-LMA's superior efficacy and reliability against state-of-the-art methods, showing lowest error and highest correlation.

Conclusions:

• The proposed U-LMA method provides accurate and reliable non-contact heart rate estimation.
• U-LMA is resistant to motion and illumination artifacts, enhancing its practical applicability.
• The method's accuracy meets clinical standards, highlighting its potential for medical use.