Abstract: A ceiling-mounted sensing unit includes (i) one or more air temperature sensors; (ii) an infrared sensor having a field of view oriented towards a floor of the room; and (iii) a microcontroller receiving readings from both the air temperature sensors and the infrared sensor, the microcontroller providing an estimated temperature at a predetermined distance above the floor of the room based on a model of the room. The model may be based on a double-exponential smoothing function obtained by matching a Kalman filter model. Alternately, the model may be itself a Kalman filter model or a machine learning trained linear model obtained using a linear regression technique, such as L2 regularization. The Kalman filter model uses a state vector that includes both the estimated temperature and a rate of change in the estimated change in temperature. The machine-trained model may be verified using a k-fold cross-validation technique.

Abstract: A ceiling-mounted sensing unit includes (i) one or more air temperature sensors; (ii) an infrared sensor having a field of view oriented towards a floor of the room; and (iii) a microcontroller receiving readings from both the air temperature sensors and the infrared sensor, the microcontroller providing an estimated temperature at a predetermined distance above the floor of the room based on a model of the room. The model may be based on a double-exponential smoothing function obtained by matching a Kalman filter model. Alternately, the model may be itself a Kalman filter model or a machine learning trained linear model obtained using a linear regression technique, such as L2 regularization. The Kalman filter model uses a state vector that includes both the estimated temperature and a rate of change in the estimated change in temperature. The machine-trained model may be verified using a k-fold cross-validation technique.