Non-Invasive AI-powered Thermal Comfort Monitoring

Differences among individuals in their thermal comfort preferences pose a significant challenge in ensuring that building environments cater to the diverse needs of occupants. Typically, thermal comfort is assessed by monitoring the surroundings using sensors placed on building and furniture surfaces, away from individuals. However, this method alone is insufficient for personalized sensing. Directly surveying individuals provides a more accurate understanding of their thermal comfort, but this approach is limited by participation rates and intermittent feedback. Therefore, there is a demand for a non-intrusive and scalable sensing solution that can continuously monitor individuals’ thermal comfort within buildings. Such a solution would enable the evaluation of whether a building meets comfort standards and facilitate more efficient climate control by avoiding unnecessary energy consumption in unoccupied spaces. In response to this need, the interdisciplinary iThCoM project has developed a non-invasive sensing framework that utilizes humans themselves as sensors to monitor the indoor environment.

This framework relies on regular RGB and infrared (IR) cameras to detect individual parameters such as sex, age group, activity, and clothing within the camera’s field of view using computer vision techniques. Additionally, limited skin temperature measurements obtained through an infrared camera serve as input to a Machine Learning-based human thermo-physiology model. This model, informed by physical principles, can then determine an individual’s thermal state in real-time, including skin temperatures across all body parts and core body temperature. In the iThCoM project, the accuracy achieved for detecting personal parameters ranged from 70% to 90%, while thermal state prediction accuracy ranged from 60% to 95%. Further refinement of this sensing framework could be achieved by expanding the input dataset and incorporating data from a larger and more diverse population.

This collaborative endeavor harnessed complementary expertise from ICE and VITA labs at EPFL. While the ICE lab concentrated on human comfort and the exchange of heat between individuals and their environment, the VITA lab specialized in computer vision and deep learning techniques. Together, our shared objective was to establish a pathway towards scalable and cost-efficient personalized comfort sensing solutions.

Project partner: Prof. Alexander Alahi, VITA lab @ EPFL

Funding: ENAC Interdisciplinary Cluster Grants 2021


  1. Rida M., Abdelfattah M., Alahi A., Khovalyg D. (2023) Toward contactless human thermal monitoring: A framework for Machine Learning-based human thermo-physiology modeling augmented with computer vision, Building and Environment, 110850, DOI:1016/j.buildenv.2023.110850
  2. Rida M., Abdelfattah M., Alahi A., Khovalyg D., Non-intrusive physiological parameters sensing for personalized human thermal comfort prediction, 18th Healthy Buildings Europe Conference, 11-14 June 2023, Aachen, Germany