A robust non-contact heart rate estimation from facial video based on a non-parametric signal extraction model
Yanrui Liu,Chunlong Xu,Lin Qi,Yongchun Li
DOI: https://doi.org/10.1016/j.bspc.2024.106186
IF: 5.1
2024-07-01
Biomedical Signal Processing and Control
Abstract:Remote photoplethysmography (rPPG) is a non-contact method to extract heart rate (HR) pulse signals from facial videos. However, challenges persist due to head motion disturbances and ambient illumination variations. To address these issues, a novel non-parametric signal extraction model is proposed for HR estimation using the webcam camera. We first select the better-quality RGB signal from the divided face regions through the ranking of signal-to-noise ratio (SNR). Then, we use blind source separation and skin reflection modeling to eliminate the effect of ambient illumination. Finally, we employ a non-parametric signal extraction model named circulant singular spectrum analysis (CiSSA), with frequency-based decomposition and energy-driven reconstruction, to extract the HR-related pulse signal effectively. To evaluate the performance of our method, we conducted the validation experiments on three datasets, including two public datasets (UBFC and PURE dataset) and one self-collected dataset (ILVA dataset). Our method outperformed significantly on all three datasets, achieving Mean Absolute Deviation (MAD) of 2.42, 1.36, and 2.84 bpm, Root Mean Square Error (RMSE) of 1.89, 2.24, and 3.58 bpm, and Pearson Correlation Coefficients (PCC) of 0.98, 0.98, and 0.93, respectively. The experimental results demonstrate that our method achieves comparable performance, particularly in scenarios involving severe head motion disturbances and variations in ambient illumination.
engineering, biomedical