Detection of wheelchair user activities using wearable sensors

Abstract

Wearable sensors are increasingly used to monitor and quantify physical activity types and levels in a real-life environment. In this project we studied the activity classification in manual wheelchair users using wearable sensors. Twenty-seven subjects performed a series of representative activities of daily living in a semi-structured setting with a wheelchair propulsion monitoring device (WPMD) attached to their upper limb and their wheelchair. The WPMD included a wheel rotation datalogger that collected wheelchair movements and an eWatch that collected tri-axial acceleration on the wrist. Features were extracted from the sensors and fed into four machine learning algorithms to classify the activities into three and four categories. The results indicated that these algorithms were able to classify these activities into three categories including self propulsion, external pushing, and sedentary activity with an accuracy of 89.4-91.9%. © 2011 Springer-Verlag.