Reconnaissance du comportement humain avec détection sans fil

Chronic diseases such as cardiovascular disorders, chronic respiratory illnesses, and diabetes are leading causes of mortality and long-term disability worldwide, posing significant challenges to healthcare systems, especially for aging populations. Traditional diagnostic and monitoring approaches rely heavily on professional personnel and centralized medical infrastructure, limiting their suitability for continuous health management. Wireless signal-based health monitoring has therefore gained increasing attention due to its non-contact, low-cost, and easily deployable nature, enabling long-term physiological sensing in daily environments. Among applications, cardiac motion, respiratory, and sleep monitoring are core research focuses due to their direct relevance to vital physiological activities.Despite progress, existing wireless health monitoring systems face three main challenges: (1) extracting fine-grained cardiac micromotions is difficult due to limited signal sensitivity or interference from respiration; (2) reliable vital sign coverage during sleep is often lost due to involuntary movements; and (3) thoraco-abdominal signal superposition impedes comprehensive respiratory assessment.This thesis addresses these gaps by proposing a model-driven framework that integrates physical signal propagation principles with the intrinsic physiological structure of the human body, improving interpretability and robustness across environments. Three systems were developed, each addressing one key challenge:Accurate Cardiac Micromotion Monitoring: A Mixed-Medium Wi-Fi Fresnel Zone Model is introduced to sense internal cardiac motion in the thoracic cavity. By exploiting Wi-Fi signal penetration and wavelength shortening in biological tissue, this physics-based approach enhances the SNR of heartbeat reflections while suppressing respiratory interference. It accurately extracts heart rate (HR), Q-T intervals, and inter-beat intervals (IBI), advancing Wi-Fi sensing from coarse rate estimation to fine-grained cardiac kinematics.Robust Sleep Vital Sign Detection: A Human Range Mapping (HRM) Model tailored for UWB radar addresses motion-induced signal dropouts during sleep. Although limb or torso movements distort the Line-of-Sight path, vital signs remain embedded in multipath components. Leveraging high range-resolution UWB and time-frequency processing, the system maintains high temporal coverage of HR and respiration rate (RR), improving reliability and longitudinal accuracy of nocturnal monitoring.Thoraco-Abdominal Signal Separation: Using mmWave FMCW radar, an Adjacent-Object Signal Superposition Model quantifies the distance-dependent contribution of thoracic and abdominal reflections. A 3D signal separation and reconstruction algorithm employing beamforming effectively decouples superimposed signals, enabling single-device monitoring of independent chest and abdominal motions, crucial for applications such as sleep apnea diagnosis and respiratory muscle assessment.The systems were validated through extensive real-world experiments with over 30 participants and hundreds of hours of data. Results demonstrate a robust, physics-grounded paradigm for comprehensive non-contact physiological monitoring, addressing the key limitations of existing wireless health sensing methods.