LAUSR

Existing heating, ventilation, and air-conditioning systems have difficulties in considering occupants’ dynamic thermal needs, thus resulting in overheating or overcooling with huge energy waste. This situation emphasizes the importance of occupant-oriented microclimate control where dynamic individual thermal comfort assessment is the key. Therefore, in this paper, a vision-based approach to estimate individual clothing insulation rate ( $$I_{\rm{cl}}$$ ) and metabolic rate (M), the two critical factors to assess personal thermal comfort level, is proposed. Specifically, with a thermal camera as the input source, a convolutional neural network (CNN) is implemented to recognize an occupant’s clothes type and activity type simultaneously. The clothes type then helps to differentiate the skin region from the clothing-covered region, allowing to calculate the skin temperature and the clothes temperature. With the two recognized types and the two computed temperatures, $$I_{\rm{cl}}$$ and M can be estimated effectively. In the experimental phase, a novel thermal dataset is introduced, which allows evaluations of the CNN-based recognizer module, the skin and clothes temperatures acquisition module, as well as the $$I_{\rm{cl}}$$ and M estimation module, proving the effectiveness and automation of the proposed approach.