iOccupancy
Chunhua Chen, Yubin Ruan, Zhongwen Liao · 2018
Accurate and timely occupancy data is critical for occupancy-driven energy efficiency in smart building management. This paper presents an on-going project iOccupancy which elaborates the practice of employing a CNN-based object detection algorithm YOLOv3 (version 3) for occupancy counting using images from surveillance cameras and online occupancy-driven HVAC control in campus classrooms. Our experiments show that counting occupants with the YOLOv3 object detector can only achieve an accuracy around 60% in large classrooms with a single camera located at the back of classrooms, indicating that more than one cameras might be needed for practical usage. Energy saving opportunities are also justified in real use cases of classrooms, comparing to fixed HVAC scheduling (e.g. using course time tables).