In this paper, we aim to clarify the effectiveness of a thermal‐based sensor, which applies image recognition technique to thermal image to measure occupancy state. Our experiment, which was conducted at a single private office over 8 days, evaluated both energy‐saving rate of occupancy control and F‐measure, which is the comprehensive metric of measurement accuracy. The thermal‐based sensor increased the F‐measure and energy‐saving rate by 2.23% and 27.91%, respectively, compared with the passive infrared sensor, which measures the occupancy state on the basis of change in infrared temperature. Compared with an image‐based sensor, which applies image recognition technique to visible image, it can provide high privacy and security preserving capability and detect occupants under completely dark environment. However, the F‐measure and energy‐saving rate were decreased by 3.13% and 19.07%, respectively. Further results and discussion are included herein toward analysis and improvement of the thermal‐based sensor. © 2022 Institute of Electrical Engineers of Japan. Published by Wiley Periodicals LLC.