Probabilistic Modelling of Gait for Robust Passive Monitoring in Daily Life

Passive monitoring in daily life may provide valuable insights into a person's health throughout the day. Wearable sensor devices play key role enabling such non-obtrusive fashion. However, data collected reflect multiple and behavior-related factors together. This creates need for structured principled analysis to produce reliable interpretable predictions that can be used support clinical diagnosis treatment. In this work we develop modelling approach free-living gait (walking) analysis. Gait is promising target because it common indicative of many different movement disorders as Parkinson's disease (PD), yet its has largely been limited experimentally controlled lab settings. To locate characterize stationary segments using accelerometers, present an unsupervised probabilistic framework designed segment signals differing non-gait patterns. We evaluate new video-referenced dataset including 25 PD patients with motor fluctuations age-matched controls, performing unscripted living activities around their own houses. Using dataset, demonstrate framework's ability detect predict medication induced based on gait. show our robust varying locations, wrist, ankle, trouser pocket lower back.