Section: Application Domains

Recognizing and predicting routine activities in smart homes

Most research on smart home systems has concentrated on techniques for recognizing context. However, many categories of service require information about likely future context. We have developed and approach that uses dynamic Bayesian networks to predict future activity and context in a smart home. Our approach extends a state-of-the-art prediction model with three contributions. First, we include sensor data through aggregation nodes, instead of limiting ourselves only to higher level context data. Second, our method uses higher order relations between activities, so that past activities can have a more meaningful impact on prediction of future activities. Third, we use a latent node that estimates the cognitive state of the occupant.