Continuous rehabilitation monitoring outside clinics is critical for long-term recovery, yet existing modalities fall short. Vision- and wearable-based systems raise privacy and compliance concerns, while RF-based sensing, despite contactless, is fundamentally motion-dependent —blind to the stillness that characterizes balance, endurance, and postural control. We observe that conventional static clutter removal not only suppresses environmental reflections but also erases the body's involuntary micro-motions, such as breathing, heartbeat, and subtle sway. Our key insight is that these micro-motions are not noise but information, encoding physiological vitality even in apparent stillness. Realizing this shift—from detecting motion to perceiving life within stillness—introduces two fundamental challenges: (C1) the echoes of these micro-motions are orders of magnitude weaker than static clutter, spectrally overlap within the near-zero Doppler region and spatially co-located with dominant reflections; and (C2) mmWave reflections are inherently sparse and geometry-agnostic, lacking the structural priors required to recover body's shape and pose across users and environments. To address these, we design mmRehab, a transformative mmWave sensing system for rehabilitation, extending radar perception beyond motion to enable physiological interpretation even when users remain still. Within mmRehab, Micro-motion Feature Extraction addresses C1 through beamforming-based spatial isolation and micro-Doppler temporal discrimination, amplifying respiration- and posture-related cues; Geometry-aware Knowledge Transfer addresses C2 via depth-guided distillation, transferring structural priors from vision to radar representations for robust generalization. Extensive experiments on both dynamic and static rehabilitation tasks show that mmRehab reduces 3D reconstruction errors by over 24% and generalizes robustly to unseen users, distances, and orientations—demonstrating the feasibility of unified radar perception for motion and micro-motion rehabilitation monitoring.
Chen et al. (Mon,) studied this question.