Real-time occupancy detection with physics-informed pattern-recognition machines based on limited CO2 and temperature sensors

An indoor occupancy estimator is proposed for the purpose of identifying the room's binary occupancy state based on limited carbon dioxide (CO2) and temperature measurements from only two available sensors. The proposed approach includes a physics-informed pattern-recognition machine (PI-PRM). Invariant features, including CO2 level, CO2 rate of change and HVAC state, are extracted by the first PI-PRM submodule based on Savitzky-Golay (S-G) filtering and other physics-informed transformations. Optimal parameterization of the feature extraction transformations is also provided based on Stein's unbiased risk estimator (SURE) and a physics-informed transformation for the HVAC operation mode. The proposed PI-PRM method is available for real-time estimation within few minutes from an occupancy change. Experimental results in an average size room, demonstrate the efficacy of our method with an accuracy of 97%. Among other applications, the proposed work can potentially be leveraged in energy efficiency, air quality improvement and emergency evacuation. Published by Elsevier B.V.

Automated summary

An indoor occupancy estimator is proposed, which can identify the room's occupancy state based on limited CO2 and temperature measurements with 97% accuracy. The method utilizes a physics-informed pattern-recognition machine for real-time estimation of occupancy changes.