Citation

Muheidat F, Tyrer HW, Muheidat F, Tyrer HW, Tyrer HW, Muheidat F. Conf. Proc. IEEE Eng. Med. Biol. Soc. 2016; 2016: 5356-5359.

Copyright

(Copyright © 2016, IEEE (Institute of Electrical and Electronics Engineers))

DOI

10.1109/EMBC.2016.7591937

PMID

28227720

Abstract

In this paper, we have enhanced smart carpet, which is a floor based personnel detector system, to detect falls using a faster but low cost processor. Our hardware front end reads 128 sensors, with sensors output a voltage due to a person walking or falling on the carpet. The processor is Jetson TK1, which provides more computing power than before. We generated a dataset with volunteers who walked and fell to test our algorithms. Data obtained allowed examining data frames (a frame is a single scan of the carpet sensors) read from the data acquisition system. We used different algorithms and techniques, and varied the windows size of number of frames (WS ≥ 1) and threshold (TH) to build our data set, which later used machine learning to help decide a fall or no fall. We then used the dataset obtained from applying a set of fall detection algorithms and the video recorded for the fall pattern experiments to train a set of classifiers using multiple test options using the Weka framework. We measured the sensitivity and specificity of the system and other metrics for intelligent detection of falls.

RESULTS showed that Computational Intelligence techniques detect falls with 96.2% accuracy and 81% sensitivity and 97.8% specificity. In addition to fall detection, we developed a database system and web applications to retain these data for years. We can display this data in realtime and for all activities in the carpet for extensive data analysis any time in the future.

Language: en