Image-based occupancy positioning system using pose-estimation model for demand-oriented ventilation

Abstract

Dynamic changes in the position, number, and activities of occupants create various indoor scenarios that affect the operation of and energy required for heating, ventilation, and air conditioning (HVAC) systems. HVAC systems can use less energy by adapting their operational parameters according to the occupant distribution. However, accurate and non-intrusive occupancy positioning systems have yet to be established for conventional HVAC and more recently developed demand-oriented ventilation (DOV) systems. Herein, a novel image-based occupancy positioning system is proposed. This system uses multiple cameras to capture images from different views and a computer to perform digital image processing and three-dimensional (3D) reconstruction. Data fusion and 3D reconstruction algorithms were developed based on skeleton key points extracted from the images with a human pose estimation model. The entire system was built and tested in a multi-function room under two different scenarios. The results showed that the system provides positioning and orientation information for a dozen of people in real time. Because the system requires an ordinary computer and conventional surveillance cameras, it can be integrated into existing video surveillance systems to avoid privacy issues. As the speed of digital image processing and computer calculations increases, the cost of the system will decrease; these characteristics provide a solid foundation for the operation of highly efficient DOV systems.