Activity Learning for Intelligent Buildings

To be considered intelligent, buildings need to operate automatically in a user-centric fashion. Intelligent buildings enable and ensure a healthy, comfortable and productive working environment for their occupants. The transition to intelligent buildings requires systems that can recognise the needs and anticipate the behaviour of the users. We foster such a transition by learning the activities that occupants perform in buildings. This is done by first recognising the activities and then predicting the time of their next occurrences. The idea rests upon an existing formalism for a building environment and applies machine learning. We implement and deploy a system in a living lab environment, showing that it can learn diverse occupant activities with high accuracy, provided simple and unobtrusive sensors.