Wearable Detection of Gait Changes in Walking Activities via Time Series Similarity Measures

Wearable technologies are increasingly employed to support continuous monitoring of human activity, enabling early detection of deviations from normal patterns in everyday movements. In this paper we study the use of time-series similarity measures to assess abnormal walking patterns, without prior knowledge of the individual, using wearable devices. We have recorded activities during normal and abnormal patterns with altered gait conditions induced by unilateral weight addition and shoe elevation. We collect data from 7 wearable devices on placed on the human being, and analyze them to center the signal on the walking action peak. This allows us to use computationally efficient similarity measures, key to be deployed on resource constrained devices. Our results show that even in short time walking segments it is possible to identify abnormal patterns, highlighting the feasibility of lightweight, real-time gait monitoring. Our results achieve >0.95 F1 score when considering multiple steps for the detection. Although the contribution of this work can be applied in several domains, we show how we leverage such information in a continuous monitoring application for telemedicine, focused on haemophilia patients.