Using dynamic occupancy patterns for improved presence detection in intelligent buildings

Presence detection is used in occupancy-based control to dynamically adjust energy-related appliances in smart buildings. Yet, practical applications typically suffer from high sensor unreliability. In our previous work, we suggested a Hidden Markov Model (HMM) for fusing information from multiple sources to better estimate the user state (presence/absence). We now extend this model and exploit information on the time-dependency of the probability of occupancy according to the time of the day. People generally have a typical working schedule, that is, occupants in an office arrive and leave every day at almost the same time. In this approach, we use our prior knowledge on office occupancy profiles to develop a time-dependent (in-homogeneous) HMM. Judging from our experiments, the algorithm shows improved performance, also, in a real-world test set-up where user presence and sensors error may not exactly follow our idealized model assumptions.