Vital Sign Sensing in Motion based on Millimeter Wave Radar

Non-contact monitoring technology can provide continuous vital sign data and help to detect the changes of physiological state in time, which is of great significance for early diagnosis and health management of diseases. For decades, however, the focus of the study was on stationary objects, as body motion clutter can easily overwhelm the subtle reflection of vital signs. This paper combines the direct RF sensing of a static example with the respiratory and heartbeat signal detection based on motion power estimation. We designed a custom deep learning model to capture the complex correlations between RF signal patterns, movement power, and vital signs. We further designed a real-time calibration and adaptive training scheme to achieve cross-topic generalization. Finally, we use FMCW radar sensor to evaluate the framework. The experimental results show that our method achieves an average HR error estimate of 4.99 bpm and 6.95 bpm on two different datasets, respectively, which is largely superior to the most advanced systems.