Human Remote Sensing In Long-Term Care Facility Using Low-Cost FMCW Radar

Abstract

This thesis studies privacy-preserving human remote sensing in Long Term Care (LTC) environments using low-cost 60 GHz FMCW radar. The main challenge we are addressing is the reliable sensing of weak or quasi-static human states in LTC environments, such as quiet occupancy, prolonged sitting or lying, and post-fall floor presence, which are clinically and operationally important yet difficult to detect with conventional methods. These scenarios are precisely the cases in which sensing reliability is most critical for resident safety, caregiver response, and building operation. The thesis, therefore, investigates how far a low-cost, low-resolution radar platform can be pushed through signal processing, machine learning methods, and simulation-driven data generation to deliver useful roomlevel awareness without relying on cameras or wearables. The work is organized around a practical deployment view rather than a single algorithmic contribution. It begins with the sensing hardware and baseline signal-processing chain, then develops methods for quasistatic occupancy detection and post-fall floor-occupancy detection, extends these ideas to imaging-style radar representations suitable for edge deployment, and finally studies simulation and digital-twin approaches for sim-to-real radar learning. Across these working stages, the unifying theme is the design of scalable, privacy-preserving, and energy-aware radar sensing methods for ambient assisted-living environments.