Exportar Publicação

Silva, B. & Marinheiro, R. N. (2021). Non-invasive monitoring with ballistocardiographic sensors for sleep management. In 2021 Telecoms Conference (ConfTELE). Leiria: IEEE.

B. Silva and R. M. Marinheiro,  "Non-invasive monitoring with ballistocardiographic sensors for sleep management", in 2021 Telecoms Conf. (ConfTELE), Leiria, IEEE, 2021

@inproceedings{silva2021_1713602391191,
	author = "Silva, B. and Marinheiro, R. N.",
	title = "Non-invasive monitoring with ballistocardiographic sensors for sleep management",
	booktitle = "2021 Telecoms Conference (ConfTELE)",
	year = "2021",
	editor = "",
	volume = "",
	number = "",
	series = "",
	doi = "10.1109/ConfTELE50222.2021.9435481",
	publisher = "IEEE",
	address = "Leiria",
	organization = "ESTG/Polytechnic of Leiria",
	url = "https://ieeexplore.ieee.org/xpl/conhome/9435429/proceeding"
}

TY  - CPAPER
TI  - Non-invasive monitoring with ballistocardiographic sensors for sleep management
T2  - 2021 Telecoms Conference (ConfTELE)
AU  - Silva, B.
AU  - Marinheiro, R. N.
PY  - 2021
DO  - 10.1109/ConfTELE50222.2021.9435481
CY  - Leiria
UR  - https://ieeexplore.ieee.org/xpl/conhome/9435429/proceeding
AB  - Sleep has an important impact on people's daily lives. A successful methodology for monitoring sleep is Polysomnography (PSG). This is an accurate and reliable approach but, unfortunately, very invasive. PSG uses expensive sensors that must be positioned by experts, what, in practice, makes its adoption only viable in hospital setups. Therefore, there is a demand for better non-invasive alternatives, such as Ballistocardiography (BCG). BCG uses cheaper sensors, easy to install and ideal for domestic use. This allows its integration in solutions that manage sleep, using mobile apps not only for presenting valuable information to users but may also for acting on the environment, through actuators, such as sound. This work uses this principle to help users to wake up smoothly. Sleep monitoring is performed with Murata SCA11H BCG external sensors. Low-pass filters have been implemented, using a sliding exponential average, for all metrics. The Random Forest algorithm was then selected for sleep phase classification, that presented the best performance when using the Weka exploration tool for learning methods. With the implemented model, it has been proved that four sleep phases are predicted. It was then possible to define a strategy for avoiding waking up alarms to be fired during deep sleep. It consists on the analysis 15 minutes prior to the alarm and, when deep sleep is detected, a relaxing sound is played. This work demonstrated that non-invasive sleep monitoring can be used to actuate on, and improve, the user environment, in a home setup with cheap sensors.
ER  -