Using the contrast within a single face heat map to assess personal thermal comfort

Abstract

An occupant-centered Heating, Ventilation, and Air Conditioning (HVAC) system accurately responds to the real thermal needs of the building occupants. To achieve this, the occupants' thermal comfort needs to be assessed in real-time with minimal inconvenience to the occupants. Whereas prior work has examined non-invasive assessment of thermal comfort using sensors aimed at a combination of body parts, this paper explores the extent to which such prediction is possible solely through non-invasive sensors aimed at the face. This is valuable because the face is least likely to be covered or occluded from sensing. The proposed method uses visual sensors to measure bio-signals related to thermal comfort. The sensor data is processed to extract local features from various parts of a face, which are used to model the thermal profile of the occupant. These features are then subject to contrast computations, which are less sensitive to the accuracy of the temperature readings. Using these features, a machine learning framework is proposed to predict the occupants’ personal thermal comfort.Data from 22 subjects recorded in an office setup with constant room temperature and fixed clothing insulation were used to test the proposed method. Three groups of subjects were exposed to different room temperatures (20 °C, 24 °C, and 28 °C respectively). The thermal comfort assessment showed that the proposed automated features can predict personal thermal comfort with 76% accuracy. We concluded by evaluating the generalization power of the proposed method using 11 additional subjects exposed to dynamic conditions, such as variable clothing insulation and room temperature.