A Bayesian approach for indoor human activity monitoring

Activity monitoring plays a crucial role in ambient living environments for assessing changes in the normal behavioral pattern of elderly people. In this paper, we present a composite action description and detection model for activity monitoring. The model accomplishes real time continuous monitoring of composite actions by detecting the transitions from one simple action to another and determining the types of those actions. It utilizes a wearable TI Chronos watch with a built-in tri-axial accelerometer for data acquisition and uses naive Bayes classifier for the classification of simple actions; walk, sit stand and lie. The unique feature of these actions is that the transition between walk, sit and lie are the most likely causes of fall event in a home environment for elderly people. The early results of an experimental study conducted for the detection of the composite actions; walk-after-sit and sit-after-lie are very encouraging in terms of detection success rates.