Citation

Ariani A, Redmond SJ, Chang D, Lovell NH. Conf. Proc. IEEE Eng. Med. Biol. Soc. 2010; 2010: 2115-2118.

Affiliation

Graduate School of Biomedical Engineering, University of New South Wales, Sydney, 2052, Australia.

Copyright

(Copyright © 2010, IEEE (Institute of Electrical and Electronics Engineers))

DOI

10.1109/IEMBS.2010.5627202

PMID

21096573

Abstract

Falls and their related injuries are a major challenge facing elderly people. One serious issue related to falls among the elderly living at home is the 'long-lie' scenario, which is the inability to get up from the floor after a fall, followed by lying on the floor for 60 minutes, or more. Several studies of accelerometer and gyroscope-based wearable falls detection devices have been cited in the literature. However, when the subject moves around at night-time, such as making a trip from the bedroom to the toilet, it is unlikely that they will remember or even feel an inclination to wear such a device. This research will investigate the potential usefulness of an unobtrusive fall detection system, based on the use of passive infrared sensors (PIRs) and pressure mats (PMs), that will detect falls automatically by recognizing unusual activity sequences in the home environment; hence, decreasing the number of subjects suffering the 'long-lie' scenario after a fall. A Java-based wireless sensor network (WSN) simulation was developed. This simulation reads the room coordinates from a residential map, a path-finding algorithm (A*) simulates the subject's movement through the residential environment, and PIR and PM sensors respond in a binary manner to the subject's movement. The falls algorithm was tested for four scenarios; one scenario including activities of daily living (ADL) and three scenarios simulating falls. The simulator generates movements for ten elderly people (5 female and 5 male; age: 50-70 years; body mass index: 25.85-26.77 kg/m(2)). A decision tree based heuristic classification model is used to analyze the data and differentiate falls events from normal activities. The sensitivity, specificity and accuracy of the algorithm are 100%, 66.67% and 90.91%, respectively, across all tested scenarios.

Language: en