Abstract: Activity recognition systems learn from sensors readings to recognise activities of the occupants in a smart home. In order to recognise activities, the sensor stream has to be segmented before any classification can be carried out. Many methods handle segmentation and recognition separately. In this study, we propose a method that can segment and recognise activity simultaneously by using a set of trained hidden Markov models and Viterbi algorithm. We evaluate our proposed method on two publicly available smart home datasets.