Super Low Resolution RF Powered Accelerometers for Alerting on Hospitalized Patient Bed Exits

TitleSuper Low Resolution RF Powered Accelerometers for Alerting on Hospitalized Patient Bed Exits
Publication TypeConference Proceedings
Year of Conference2019
AuthorsChesser, M., A. Jayatilaka, R. Visvanathan, C. Fumeaux, A. Sample, and D. C. Ranasinghe
Conference NameIEEE International Conference on Pervasive Computing and Communications (Percom 2019)
Date Published2019
Abstract

Falls have serious consequences and are prevalent inacute hospitals and nursing homes caring for older people. Mostfalls occur in bedrooms and near the bed. Technological interven-tions to mitigate the risk of falling aim to automatically monitorbed-exit events and subsequently alert healthcare personnel toprovide timely supervisions. We observe that frequency-domaininformation related to patient activities exist predominantly invery low frequencies. Therefore, we recognise the potential toemploy a low resolution acceleration sensing modality in contrastto powering and sensing with a conventional MEMS (MicroElectro Mechanical System) accelerometer. Consequently, weinvestigate a batteryless sensing modality with low cost wirelesslypowered Radio Frequency Identification (RFID) technology withthe potential for convenient integration intoclothing, such ashospital gowns. We design and build apassiveaccelerometer-based RFID sensor embodiment—ID-Sensor—for our study. Thesensor design allows deriving ultra low resolution accelerationdata from the rate of change of unique RFID tag identifiersin accordance with the movement of a patient’s upper body.We investigate two convolutional neural network architecturesfor learning from rawRFID-onlydata streams and compareperformance with a traditional shallow classifier with engineeredfeatures. We evaluate performance with 23 hospitalized olderpatients. We demonstrate, for the first time and to the best ofknowledge, that: i) the low resolution acceleration data embeddedin the RF poweredID-Sensordata stream can provide a practica-ble method for activity recognition; and ii) highly discriminativefeatures can be efficiently learned from the raw RFID-only datastream using a fully convolutional network architecture.

DOI10.1109/PERCOM.2019.8767398