Real-world Gyroscope-based Gait Event Detection and Gait Feature Extraction

Falls in older adults are a major clinical problem often resulting in serious injury. The costly nature of clinic-based testing for the propensity of falling and a move towards homebased care and monitoring of older adults has led to research in wearable sensing technologies for identifying fall-related parameters from activities of daily living. This paper discusses the development of two algorithms for identifying periods of walking (gait events) and extracting characteristic patterns for each gait event (gait features) with a view to identifying the propensity to fall in older adults. In this paper, we present an evaluation of the algorithms involving a small real-world dataset collected from healthy adults in an uncontrolled environment. 92.5% of gait events were extracted from lower leg gyroscope data from 5 healthy adults (total duration of 33 hours) and over 95% of the gait characteristic points were identified in this data. A user interface to aid clinicians review gait features from walking events captured over multiple days is also proposed. The work presents initial steps in the development of a platform for monitoring patients within their daily routine in uncontrolled environments to inform clinical decision-making related to falls