Office Multi-Occupancy Detection using BLE Beacons and Power Meters

Indoor occupancy provides information about human occupation in the closed space, most notably, office and residential buildings. This information is useful in dwindling unnecessary energy usage, such as consumption in unoccupied spaces or energy-wasting due to unnecessarily active appliances. We present an empirical experiment on office occupancy detection using simple office sensors. We choose generic power meters and mobile phones. First, we classify beacon signals received by mobile phones into a room location. A workspace map is assumed to be available to facilitate the mapping between room locations and the occupancy state of users' workspace. Second, we infer the individual occupancy state utilizing the aggregated electricity consumption of occupant-related devices (i.e., monitors) in shared offices. The later solution helps to keep costs and intrusiveness level low compared to deploying a power meter for each device or user. We experiment in an work environment with two shared offices, a personal office, and a social corner involving five volunteers. Given the acquired data, three techniques based on machine learning, optimization, and probabilistic approach are implemented and compared to evaluate their performance. The results indicate that localization and occupancy based on beaconing works best for three of the five volunteers, reaching 95% F-measure. Further findings shows that occupancy inference based on the aggregated power consumption performs well for the four volunteers when using Decision Tree classification, reaching more than 90% F-measure. Our effort on the fusion of two modalities gives a positive result for all five volunteers, ranging from 92% to 99% F-measure.