A wavelet-based decomposition method for a robust extraction of pulse rate from video recordings

Area of Science:

• Biomedical Engineering
• Signal Processing
• Optical Sensing

Background:

• Remote photoplethysmography (rPPG) enables non-contact pulse rate (PR) assessment from videos.
• Real-time rPPG applications require robust algorithms to eliminate signal distortions.

Purpose of the Study:

• To enhance rPPG accuracy by improving distortion elimination.
• To introduce and evaluate the Continuous-Wavelet-Transform-based Sub-Band rPPG (SB-CWT) method.

Main Methods:

• Increased dimensionality of RGB video signals using generalized Morse wavelet transform decomposition.
• Evaluated the SB-CWT method on the MMSE-HR database with 101 facial videos and reference blood volume pulse (BVP) signals.
• Compared SB-CWT performance against the state-of-the-art Sub-band rPPG (SB) method.

Main Results:

• SB-CWT achieved higher median signal-to-noise ratios (SNR) (6.63–10.39 dB) compared to SB (4.23–6.24 dB).
• SB-CWT showed superior agreement between estimated and reference PRs (R²: 0.81–0.91) versus SB (R²: 0.41–0.47).
• Mean differences in PR estimation were significantly lower for SB-CWT (1.82–5.37 BPM) than for SB (18.80–22.18 BPM).

Conclusions:

• The proposed SB-CWT method significantly outperforms the SB method.
• SB-CWT demonstrates improved SNR and better agreement for PR estimation from RGB videos.
• This advancement is crucial for real-time, non-contact vital sign monitoring.