Multi-Modal, Remote Breathing Monitor

Area of Science:

• Multimodal sensing
• Biomedical engineering
• Signal processing

Background:

• Breathing rate monitoring is crucial for applications such as infant care, sleep studies, and elder monitoring.
• Existing methods may have limitations in accuracy or applicability across diverse scenarios.

Purpose of the Study:

• To develop and validate a multimodal system for accurate human breathing rate estimation.
• To fuse data from vision-based (RGBD) and radio frequency (UWB radar) sensors for enhanced performance.

Main Methods:

• Utilized Intel® RealSense™ RGB-D sensor for visual data and ultra-wideband (UWB) radar for RF data.
• Employed Lucas-Kanade optical flow for tracking body features and depth, and processed radar signals for breathing patterns.
• Applied a generalized likelihood ratio test for harmonic detection and a reformed Pisarenko algorithm for spectral estimation, fusing results via maximum likelihood.

Main Results:

• The multimodal system successfully tracked breathing patterns from both vision and RF modalities.
• A maximum error of only 0.5 BPM was reported when compared to the true breathing rate across 14 human subjects.

Conclusions:

• The fusion of vision-based and RF-based modalities provides a robust and highly accurate method for breathing rate estimation.
• This approach holds significant potential for non-invasive monitoring in various healthcare and personal safety applications.