AI-based Space Occupancy Estimation Using Environmental Sensor Data

Building energy management - as a tool to effect day-to-day energy savings - is influenced by space occupancy levels. The use of video images to estimate occupancy levels has privacy concerns and is cost ineffective. In this paper, a Deep Learning (DL) based approach is proposed to estimate the number of people in a given space using environmental sensor data. Five environmental factors are considered to train and test the proposed model. The input data are pre-processed with $Z_{scor\mathrm{e}}$ normalization for better performance of the model. Further, the proposed method is compared with Long Short Term Memory (LSTM) with higher $F_{1}$ score. In addition, to improve the estimation accuracy of space occupancy, one hot encoding is done for output data. The proposed model estimates the number of students in classrooms with high accuracy in a cost-effective manner while maintaining privacy. The application of the proposed approach is to improve energy efficiency by utilising the estimated headcount information.