PhD Thesis Defense: Md. Farhan Tasnim Oshim, Towards High-Fidelity Motion Characterization using Radar Vibrometry: Applications in Vital Sign Monitoring and Human-Object Interaction
Content
Speaker:
Abstract:
Radar sensing has emerged as a powerful modality for contactless perception due to its robustness to lighting conditions, ability to penetrate obstacles, and inherent privacy advantages over cameras. Leveraging radar vibrometry, this dissertation develops a comprehensive framework for high-fidelity motion characterization, addressing challenges of sensitivity, robustness, and privacy. While traditional radar excels at detecting coarse movements, this work emphasizes extracting finer-scale motion to enable advanced applications in healthcare, human-object interaction, and intelligent, privacy-aware environments.
The research first addresses the challenges of contactless vital sign monitoring in clinical settings, where conventional measurement techniques are often obtrusive or inconsistent. By applying Eulerian phase-based motion magnification to radar signals, subtle physiological vibrations are amplified, enabling robust and accurate estimation of respiratory and cardiac activity in hospitalized patients, including those with respiratory distress.
Beyond biological sensing, this research explores computationally enhanced interaction and object sensing. MechanoBeat uses 3D-printed harmonic oscillators as battery-free mechanical tags, generating unique radar-detectable vibrations to infer interactions with everyday objects. Building on this concept, MetaScatter introduces computationally optimized meta-reflectors that shape objects’ radar reflections, allowing passive identification based on their radar signatures. To overcome hardware limitations and sparse, noisy measurements, a NeRF-enabled Analysis-Through-Synthesis (ATS) framework is introduced that integrates UWB radar wave propagation, object reflection characteristics, and scene priors to reconstruct high-resolution 2D radar images of small objects.
Finally, the Anti-Sensing framework addresses privacy concerns by introducing a novel defense mechanism to prevent unauthorized radar-based sensing. The approach leverages wearable devices that generate gradient-optimized oscillatory perturbations, mimicking natural cardiac motion to mislead unauthorized heart rate estimations.
Collectively, these contributions establish radar vibrometry as a robust, privacy-preserving modality for intelligent, responsive, and secure environments, spanning healthcare, human-object interaction, and ambient sensing applications.
Advisor:
Tauhidur Rahman