Real-time space occupancy sensing and human motion analysis using deep learning for indoor air quality control

This study proposed a novel indoor air quality control methodology that uses occupancy sensing and motion recognition techniques in combination with human motion analysis. The automated occupancy sensing systems that are most prevalent today analyze either environmental conditions (i.e., room temperature or data from entities in the room, such as states of electronic devices) to make human occupancy predictions. Since little emphasis is placed on observing humans directly, the estimations of these sensing systems are often inaccurate. Erroneous occupancy estimation leads to poor control of building resources such as HVAC (Heating, Ventilation, and Air-conditioning) and lighting systems. To address this issue, the paper puts forth a lean, vision-based system that estimates the number of occupants and recognizes their activities using the stacked history of unconstrained non-deterministic human movements over transient intervals. The system implements a multi-stream deep neural network to identify human activities and uses the YOLO V3 deep neural network for object detection to estimate occupancy count in a room. The study uses a publicly available action recognition dataset - NADA - to train the neural networks and experiment with a variety of video classification techniques to achieve higher accuracies.