visionSCG turns ordinary cameras, including smartphones, into tools for measuring the heart's mechanical activity without touching the body. Each heartbeat produces tiny vibrations of the chest wall, usually recorded as the seismocardiogram (SCG) with an accelerometer attached to the chest. We extract these vibrations from video instead: first by tracking small stickers on the chest, then without any markers using optical flow, and most recently as multichannel SCG maps across the chest, sharpened with deep learning. Each method is validated in human subject studies against simultaneous accelerometer and ECG recordings.

We also estimate heart rate and heart rate variability from videos of the neck, a more privacy-friendly alternative to the face, and develop signal processing and deep learning methods for SCG, such as identifying heart valve events without an ECG and reconstructing three-axis SCG from two axes.

Objective

Develop contactless methods for the monitoring of cardiovascular activity based on cardiovascular-induced vibrations of the body surface

Skillsets

• Human subject studies with synchronized video, accelerometer, ECG, and PPG recordings
• Computer vision: template tracking and optical flow (Lucas–Kanade, Gunnar–Farneback)
• Remote photoplethysmography (rPPG)
• Signal processing of seismocardiograms, including ECG-free detection of heart valve events
• Deep learning (LSTM networks, super-resolution)
• Method validation (Bland–Altman, correlation, and error analysis)

Researchers

Dr. Amirtahà Taebi

Principal Investigator