PLCount: A Probabilistic Fusion Algorithm for Accurately Estimating Occupancy from 3D Camera Counts

The number of occupants in a whole building, a zone or a room is an important parameter when improving the energy efficiency of a building. Using camera-based 3D or thermal sensors to detect passing of count lines are becoming more and more common. However, such sensors are not perfect and errors add up over time. In this paper we present the PLCount algorithm to accurately estimate total building, zone or room counts by fusing data from multiple count lines. The algorithm applies probabilistic reasoning together with occupancy constraints to accurately estimate the total number of occupants. We evaluate the algorithm on two data-sets from a small and a large office building. The evaluation shows a considerably lower RMSE compared to the raw counts and naive correction approach with up to 86% and 70% error reduction respectively and similarity analysis of consecutive weeks demonstrate the stability of the algorithm over time. We also demonstrate the use of the data for analyzing the energy consumption of a building. By presenting more accurate algorithms for estimating total occupancy we hope to enable buildings to better serve the actual number of occupants to improve comfort and energy efficiency of buildings.