Towards unsupervised learning of thermal comfort using infrared thermography

Maintaining thermal comfort in built environments is important for occupant health, well-being, and productivity, and also for efficient HVAC system operations. Most of the existing personal thermal comfort learning methods require occupants to provide feedback via a survey to label the monitored environmental or physiological conditions in order to train the prediction models. Accuracy of these models usually drops after the training process as personal thermal comfort is dynamic and changes over time due to climatic variations and/or acclimation. In this paper, we present a hidden Markov model (HMM) based learning method to capture personal thermal comfort using infrared thermography of the human face. We chose human face since its blood vessels has a higher density and it is not covered while performing regular activities in built environments. The learning algorithm has 3 hidden states (i.e., uncomfortably warm, comfortable, uncomfortably cool) and uses discretization for forming the observed states from the continuous infrared measurements. The approach can potentially be used for continuous monitoring of thermal comfort to capture the variations over time. We tested and validated the method in a four-day long experiment with 10 subjects and demonstrated an accuracy of 82.8% for predicting uncomfortable conditions.