Beat-to-beat heart rate estimation fusing multimodal video and sensor data.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Wearable Technology

Background:

• Unobtrusive biosignal measurement accuracy is often limited compared to clinical methods.
• Sensor fusion can enhance data reliability by integrating information from multiple sources.
• Existing unobtrusive methods struggle with coverage and precision.

Purpose of the Study:

• To improve the accuracy and coverage of unobtrusively measured biosignals.
• To demonstrate the efficacy of sensor fusion using readily available hardware like webcams and chairs.
• To validate a novel method against a gold standard clinical measurement (ECG).

Main Methods:

• Extracting skin color variation and head motion from webcam video streams.
• Acquiring ballistocardiographic (BCG) signals using a sensor integrated into a chair.
• Employing a Bayesian approach for sensor fusion of the extracted and BCG signals.
• Comparing fused signal accuracy and coverage against an electrocardiography (ECG) reference.

Main Results:

• Achieved mean absolute beat-to-beat estimation error below 25 milliseconds.
• Obtained average signal coverage exceeding 90% compared to the ECG reference.
• Demonstrated successful fusion of video-derived and BCG signals for enhanced biosignal monitoring.

Conclusions:

• Sensor fusion of unobtrusive modalities significantly enhances biosignal measurement accuracy and coverage.
• Webcam and chair-based sensors offer a viable, low-cost alternative for improved physiological monitoring.
• The Bayesian fusion approach provides a robust method for integrating multi-channel biosignals.