Motion-robust heartbeat reconstruction from radar signals via transfer learning
Author(s)
Type
Conference Paper
Abstract
Robust heartbeat monitoring with radar in real-world settings remains challenging because random body movements can dominate cardiac micro motion and introduce phase discontinuities. This work formulates radar heart monitoring as a waveform reconstruction problem and proposes a signal processing and machine learning framework that maps radar data to a heart motion signal suitable for peak-to-peak interval analysis. To improve generalisation under random body movement scenario, the machine learning model is initialised via transfer learning on a large paired radar and photoplethysmography (PPG) sleep dataset and then adapted to the challenge domain. Experiments on the IEEE AESS random body movement (RBM) benchmark demonstrate that the proposed approach improves both average heart-rate and variability estimation under motion, achieving a heart rate root mean square error (RMSE) of 1.0 bpm, while maintaining stable performance across stationary, deep-breathing, and RBM conditions. These results suggest that transfer learning based denoising with transferable cardiac priors offers a practical route towards deployable heartbeat reconstruction in motion-contaminated radar sensing.
Date Issued
2026-05-01
Date Acceptance
2026-09-01
Citation
2026 IEEE Radar Conference (RadarConf'26), 2026
ISBN
979-8-3315-6022-5
ISSN
2375-5318
Publisher
IEEE
Journal / Book Title
2026 IEEE Radar Conference (RadarConf'26)
Copyright Statement
Copyright © 2026, IEEE. This is the author’s accepted manuscript made available under a CC-BY licence in accordance with Imperial’s Research Publications Open Access policy (www.imperial.ac.uk/oa-policy)
License URL
Identifier
https://ieeexplore.ieee.org/document/11675438
Source
2026 IEEE Radar Conference (RadarConf'26)
Publication Status
Published
Start Date
2026-05-11
Finish Date
2026-05-15
Coverage Spatial
Phoenix, AZ, USA
Date Publish Online
2026-09-04
