Development of a Human Pose Classifying Model to Estimate the Metabolic Rate of Occupant
Eun Ji Choi, Bo Rang Park, Young Jae Choi, Jin Woo Moon · KIEAE Journal · 2018
Purpose: The aim of this study is to develope a human pose classifying model from the image data for estimating occupant metabolic rate (MET) which will be used for the optimal control of indoor PMV (predictive mean vote) conditions. The deep learning artificial neural network was applied for developing the classifying model. Methods : For developing the classifying model, the fundamental data set about occupant diverse indoor activities was collected for training the model and testing its performance. The model structure and parameter values were determined through the optimization process and the k-fold Cross Validation was conducted for the model training. The performance evaluation of the classifying model was accomplished using the accuracy of the MET from the pose classification. Results : The optimized model, which was set to have 1 hidden layer and 200 hidden neurons 200, presented 88.01% accuracy to properly estimate the MET from the classified occupant pose. Based on this accuracy, the model showed a potential to be applied to estimating the MET and indoor PMV conditions.