View article

This article introduces a methodological approach for analysing time series data from multiple sensors in order to estimate home occupancy. The approach combines the Dempster-Shafer theory, which allows the fusion of ‘evidence’ from multiple sensors, with the Hidden Markov Model. The procedure addresses some of the practicalities of occupancy estimation including the blind estimation of sensor distributions during unoccupied and occupied states, and issues of occupancy inference when some sensors have missing data. The approach is applied to preliminary data from a residential family home on the North Coast of Scotland. Features derived from sensors that monitored electrical power, dew point temperature and indoor CO₂ concentration were fused and the Hidden Markov Model applied to predict the occupancy profile. The approach shown is able to predict daytime occupancy, while effectively …