Thermography-based assessment of mean radiant temperature and occupancy in healthcare facilities

Due to its high demands regarding indoor environmental conditions, healthcare facilities are associated with high energy consumption. To move forward towards more demand driven and energy reduced conditioning, information on occupancy and temperature boundary conditions are crucial. Thermography-based systems enable data acquisition regarding both aspects in high local resolution. In this publication, we propose a thermography system that may be used for monitoring of rooms in healthcare facilities. It is set up using a 160 x 120 px thermography sensor and Raspberry Pi computer for data acquisition and processing. The sensors are mounted on walls to capture the inside of the room including patients, staff, and visitors. We evaluate the mean radiant temperature based on the individual inner surfaces of the room. The algorithm aggregates wall, floor and ceiling surface temperatures within the field of view of the sensor. For occupancy estimation inside the room, we apply a convolutional neural network (CNN). It is based on a pre-trained network and retrained using a partial dataset collected during the field study. To improve robustness of the algorithm several data pre-processing steps are conducted, that include image filters and redundancy testing. The system is evaluated based on data collected in a field study conducted inside MHH Hospital in Hannover, Germany. Several patients’ rooms and a staff room are monitored over a period of 6 weeks, with the goal of evaluating indoor environmental data. The measurement period is inside the heating period in winter and different room layouts are considered. For reference, an indoor environmental quality measurement device is used to simultaneously measure air temperature, globe temperature and other IEQ parameters. Measured data of the reference system agree well with the thermography system. Deviations between both are less than 1 K in radiant temperature for most scenarios and measurement setups. Estimated occupancy is compared to a ground truth derived from manual processing of the captured thermography data. Finally, results of the field study are discussed together with the systems advantages and limitations with regard to privacy considerations.