Floor Based Sensors Walk Identification System Using Dynamic Time Warping with Cloudlet Support
Robert Hughes, Fadi Muheidat, Mason Lee, Lo’ai Tawalbeh · 2019
Current advances in sensor design technologies and computing power (computational and artificial intelligence) have made it possible to build smart assistive living systems that can improve the lives of people. Because older adults want to age in place at home, there is a need to monitor their health status, detect emergency situations, and notify health care providers. We have improved a floor based monitoring system, which we call the smart carpet, originally to detect falls, but we can take advantage of the continual 24/7 monitoring capability to get important information on gait, fall detection, counting the number of people traversing the carpet and studied the waveform for useful information. Recently, we studied the characteristics of the waveform of the scavenged signal from the sensors and used computational intelligence and feature extractions and classifications to separate people. In this paper, we used Dynamic Time Warping (DTW) to help improve on walk identification, compared with the MFCC feature extraction methods. Results showed that our system identifies walks using a dynamic time warping algorithm and KNN classifier with 86% precision, 76% recall, 81% accuracy. We also present a cooperative cloudlet mobile computing model for eldercare and medical applications where the decisions are very time sensitive. The sensors data will be sent to the nearest cloudlet for analysis and extracting real-time decisions in minimal delay. Users can obtain these results and make decisions by accessing the cloud through their mobile devices and in a real time manner.