Behaviour recognition from sensory streams in smart environments

Abstract

One application of smart homes is to take sensor activations from a variety of sensors around the house and use them to recognise the particular behaviours of the inhabitants. This can be useful for monitoring of the elderly or cognitively impaired, amongst other applications. Since the behaviours themselves are not directly observed, only the observations by sensors, it is common to build a probabilistic model of how behaviours arise from these observations, for example in the form of a Hidden Markov Model (HMM). In this paper we present a method of selecting which of a set of trained HMMs best matches the current observations, together with experiments showing that it can reliably detect and segment the sensor stream into behaviours. We demonstrate our algorithm on real sensor data obtained from the MIT PlaceLab. The results show a significant improvement in the recognition accuracy over other approaches. © Springer-Verlag Berlin Heidelberg 2009.