Deep-vision-based metabolic rate and clothing insulation estimation for occupant-centric control

The metabolic rate (MET) and clothing insulation (CLO), which are personal characteristics, are generally difficult to estimate automatically. In recent years, a combination of deep learning and computer vision (hereafter referred to as “deep vision”) has enabled real-time estimation of these characteristics. Although many studies have been conducted on the topic, practical methods for simultaneous estimation of MET and CLO and building control strategies based on these characteristics have not been sufficiently examined. This study proposes a preliminary method for classifying two activities and two clothing ensembles along with a real-time estimation of MET and CLO. In addition, a comfort temperature control strategy based on MET and CLO in a purpose-built chamber is implemented. The proposed method estimated the activities and clothing ensembles of five subjects with an accuracy of 97%, and MET and CLO with an accuracy of 100% each in the hypothetical scenario. The comfort control successfully adjusted the set-point temperature of the air conditioner according to changes in MET and CLO. Moreover, the comfort control strategy maintained the thermal sensation votes of eight subjects constant irrespective of changes in MET and CLO, and the proportion of votes representing no thermal change increased by 17% compared to that of the fixed set-point control strategy. • Deep-vision-based MET and CLO estimation method was proposed. • MET and CLO were accurately and reliably estimated in a built environment. • Comfort control was implemented in the climate chamber considering MET and CLO. • Comfort control was effective in improving subjects' TSV and TP.