返回Google圖書搜尋

Radio-frequency Sensor Applications for Assisted Living

出版	ProQuest Information and Learning, 2020
URL	http://books.google.com.hk/books?id=lm5M0AEACAAJ&hl=&source=gbs_api

註釋The demographic and socio-economic trends have signaled an overwhelming need for improved assisted living facilities with non-intrusive technologies that address prominent user needs of wellness by continuous vital-sign sensing and safety and accessibility by occupant-detection. This home-centric approach can greatly improve quality-of-life for the elderly suffering from diseases like sleep apnea and chronic obstructive pulmonary disorder (COPD) by long-term continuous monitoring. Moreover, with the ability to track occupant presence, location, and posture, the response time can be shorter in case of emergencies like fall-events.To solve the problem of wellness monitoring, this work focuses on using an over-clothing wearable radio-frequency (RF) sensor to accurately measure heartbeat and respiration. While ambient and wearable chest-motion monitoring sensors exist, they only measure the weak surface motion. This work uses near-field coherent sensing to strongly couple to internal dielectric boundary motion. As the choice of antenna placement and sensor setup has a significant impact on the user comfort level and signal quality, different sensor design variations are considered. Both thorax and abdominal respiratory patterns are measured independently to monitor paradoxical respiration during obstructive apnea. A calibration-based approach is used to estimate respiratory volume, a key parameter to estimate respiratory effort, and results are presented for different simulated breathing disorders. Finally, results are validated under posture and gender variations in a study conducted on 20 participants, showing a high correlation of presented sensing setup with the reference devices: heart rate (r = 0.95), respiratory rate (r = 0.93), and respiratory volume (r = 0.84), along with high accuracies of 96% and 83% for simulated central and obstructive sleep apnea detection respectively. The work has been extended for attention vs relaxation state classification resulting in a satisfactory accuracy of 93% on the unseen test subjects. While vital sign monitoring is an active task, which has strong advantages by a wearable or on-the-furniture sensor approach, safety and accessibility need to be embedded in the ambient, where it can monitor occupant presence to trigger responses even if the occupant is device-free. As ambient sensors need to have sufficient observation coverage in arbitrary room layout and furnishing, ambient passive radio frequency identification (RFID) tags are chosen to provide scalability cost-effectively and reliably. However, the indoor RF signal suffers from heavy multipath and unknown phase offsets due to cables and reader transceiver circuitry. In this work, a novel background calibration algorithm is presented that works well under significant background clutter. A linearized inverse model is used to generate an 'RF image' based on the occupant reflectivity and absorption, using novel sparsity approximation algorithms, from which occupant count and location can be extracted. As the number of occupants increases, the approximations fail and counting becomes challenging. In this scenario, a deep-learning-based solution is presented that can count with good accuracy of 82%, with learning transference to different rooms and setup variations, significantly reducing the cost overhead from training data collection.