Home // AMBIENT 2018, The Eighth International Conference on Ambient Computing, Applications, Services and Technologies // View article

Daily Life Monitoring System with Behavior Pattern Recognition Using Ambient Sensors

Authors:
Hirokazu Madokoro
Nobuhiro Shimoi
Kazuhito Sato

Keywords: ambient sensors; home agent; life monitoring; quality of life; machine learning; random forest.

Abstract:
This paper presents a novel life monitoring system using a home agent and ambient sensors as invisible sensors that fit living circumstances with consideration of privacy and Quality of Life (QoL) for achieving autonomous monitoring in daily life. The home agent has a key tag sensor, a human detection sensor, and a remote control sensor for detecting the states of a subject, such as going out (Out) or being at home (Home). The ambient sensors consist of a pad sensor installed in a bed sheet, a triaxial accelerometer inserted in a pillow, a human detection sensor installed near an entrance door, and a piezoelectric sensor installed near a refrigerator. The state of Home or sleeping on a bed (Sleep) is detected using ambient sensors that measure living behavior patterns in real time. As a preliminary experiment aimed at monitoring various life patterns of elderly people, we conducted a monitoring experiment during two months for four university students subjects in their 20s. For this system, sensor signals were stored in a server via a wireless router for visualization on a monitoring computer terminal in real time. We developed a method of recognizing three major life patterns (Out, Home, and Sleep) using machine learning which uses eight algorithms. To evaluate of recognition accuracy, we collected handwritten daily records from the respective subjects used for correct behavior datasets as ground truth. Experimentally obtained results revealed that the mean recognition accuracy was 83.61% for the first half of the monitoring experiment during one month with one-minute downsampled signal intervals. In the last half of the monitoring experiment, data acquisition was interrupted because of a failure of the home agent. We continued the evaluation experiment of life patterns with two delimited periods, which indicated recognition accuracies of 92.53% for 18 days and 93.85% for 27 days.

Pages: 11 to 16

Copyright: Copyright (c) IARIA, 2018

Publication date: November 18, 2018

Published in: conference

ISSN: 2326-9324

ISBN: 978-1-61208-679-8

Location: Athens, Greece

Dates: from November 18, 2018 to November 22, 2018