Non-invasive thermal sensation recognition based on human behavior postures in office environment

As an important part of intelligent environmental control, human thermal sensation prediction can not only improve human thermal comfort and work efficiency, but also help promote building energy efficiency. In this paper, a database of thermal adaptive actions is established according to thermal response of human body in the office environment. A key frame extraction algorithm is proposed based on the temporal and spatial characteristics of human skeleton data. The feature data of key frames is input into the bi-directional long short-term memory (BI-LSTM) neural network model for thermal comfort state recognition. Results show that for seven-category thermal sensation recognition, the accuracy of proposed model achieves 86.9%, which demonstrates the feasibility of the camera-based non-invasive identification method.