Bayesian-Based Weighted Scheme for Wireless Vital Sign Estimation in Small Data Conditions

Vital signs are fundamental in diagnosing diseases and determining human health conditions. Considering the importance of developing a non-contact solution for vital sign estimation during an epidemic outbreak, extensive research based on millimeter-wave (mmWave) frequency-modulated continuous-wave (FMCW) radar has been proposed. Current well-known deep learning-based methods perform well, but have been challenged when the number of training samples is limited, leading to risks of overfitting and poor generalization. In this study, a Bayesian-based weighted scheme is proposed as a non-parametric data driven solution to provide satisfactory performance under small data conditions. The proposed scheme enhances adaptability to diverse propagation environments with limited data. This is achieved by leveraging non-parametric statistical calculations to reduce data dependency and by employing a data driven weighting mechanism, which dynamically assigns greater importance to more accurate estimations. The experimental results first offer foundational insights into the proposed scheme in a single-person scenario, followed by extensive evaluation using a publicly available dataset of multi-person scenarios at long detection distances. The proposed Bayesian-based weighted scheme achieves convincing performance in terms of accuracy and heart rate variability (HRV) monitoring while demonstrating superior data efficiency.