Abstract
This study aims to propose a pose classification model using indoor occupant images. For developing the intelligent and automated model, a deep learning neural network was employed. Indoor posture images and joint coordinate data were collected and used to conduct the training and optimization of the model. The output of the trained model is the occupant pose of the sedentary activities in the indoor space. The performance of the developed model was evaluated for two different indoor environments: home and office. Using the metabolic rates corresponding to the classified poses, the model accuracy was compared with that of the conventional method, which considered the fixed activity. The result showed that the accuracy was improved by as much as 73.96% and 55.26% in home and office, respectively. Thus, the potential of the pose classification model was verified for providing a more comfortable and personalized thermal environment to the occupant.
Funding Information
  • Korea Institute of Energy Technology Evaluation and Planning (20182010600010)
  • Chung-Ang University (Graduate Research Scholarship in 2019)