Millimeter Wave Radar Calibration for Heart Rate Estimation using Bayesian Neural Networks.

Human monitoring for better life quality, especially for elderly population, is one of the main concerns in today's research. The estimation of vital signs is an active field of study, and multiple approaches have been proposed. The use of millimetre wave (mmWave) radars is of interest because of their non-contact monitoring capabilities and high accuracy. However, using this technology poses challenges and compensating for them is crucial for a proper deployment. To address proper calibration, absorbing different sources of misalignment, this paper proposes a Bayesian Neural Network to calibrate mmWave radar heart rate estimates with respect to a certified medical device. This solution is capable of predicting a 95% confidence interval around the estimated value for each input, and its reliability is assessed with a statistical hypothesis test on a separate dataset. The experiments have been performed with a frequency-modulated continuous wave radar device, centered at 77 GHz. For the test dataset, the average error achieved was 1.12 bpm.