Evaluation of wearable sensor tag data segmentation approaches for real time activity classification in elderly

Abstract

The development of human activity monitoring has allowed the creation of multiple applications, among them is the recognition of high falls risk activities of older people for the mitigation of falls occurrences. In this study, we apply a graphical model based classification technique (conditional random field) to evaluate various sliding window based techniques for the real time prediction of activities in older subjects wearing a passive (batteryless) sensor enabled RFID tag. The system achieved maximum overall real time activity prediction accuracy of 95% using a time weighted windowing technique to aggregate contextual information to input sensor data.