Thermal Comfort and Productivity under Air Conditioning Control Based on Metabolic Rate Estimation from Thermal Images

¹ Department of Architecture and Building Engineering, School of Environment and Society, Institute of Science Tokyo, Tokyo, 1528550, Japan
² AI Solution Department, Azbil Corporation, Kanagawa, 2518522, Japan

^* Corresponding author: This email address is being protected from spambots. You need JavaScript enabled to view it.

Abstract

Conventional air conditioning (AC) control does not account for human factors such as metabolic rate—a key factor affecting thermal sensation—and therefore cannot meet individual thermal comfort needs. However, recent advances in thermal imaging sensors and AI-based image analysis have opened up the possibility of AC control that more closely aligns with individuals' thermal states. Therefore, we proposed an AC control based on metabolic rate estimation using thermal images and conducted a subject experiment comparing it with conventional AC control. Since continuous measurement of metabolic rate is difficult, we estimated it non-invasively using average surface temperature obtained from thermal images. Human body regions were extracted in real time through AI-based image analysis, and their surface temperatures were used to estimate individual metabolic rates based on the relationship between surface temperature and metabolism. The estimated values were then applied to AC control using Predicted Mean Vote (PMV) as the control target. The subject experiment was conducted in September 2024 during the summer season in the climate chamber. Six participants took part in each session, with a total of 12 participants across two sessions. Three cases were tested in the experiment. In Case 1, the room temperature was kept constant at 26°C, representing conventional AC control. In Case 2 (room temperature control to prevent overcooling), the room temperature was lowered when the minimum PMV among the six participants exceeded +0.5. In Case 3 (room temperature and airflow control), building upon Case 2, individual airflow was added using personal fans to adjust each participant's PMV closer to ±0. For each case, participants performed tasks for 6 hours per day—2 hours in the morning and 4 hours in the afternoon. The results showed that thermal comfort votes were significantly higher and requests for room temperature adjustment were significantly fewer in Case 3 compared to Cases 1 and 2. In Case 3, alertness—an indicator of subjective productivity—was significantly higher, accompanied by improved objective performance in the calculation task. These findings indicate that AC control using thermal images is more human-centered and can improve thermal experience while enhancing productivity.