Building occupancy detection using thermal imaging based edge computing

Document Type

Article

Publication Date

1-1-2026

Abstract

As energy costs continue to rise, energy efficiency within buildings is becoming increasingly important to reduce energy costs for consumers and to improve the competitiveness of businesses. Occupancy detection has been proven to be an efficient way to improve building energy performance by smartly enabling the control of the heating, ventilating, air-conditioning (HVAC), and lighting systems. Several devices have been developed and practically employed in buildings to enable occupancy detection; however, there are still major challenges in low accuracy, required motion, and/or inability to detect focused areas. This paper presents an edge computing thermal imaging system that can precisely detect the presence and count the number of people in a focused area in real-time. A convolutional neural network (CNN) with fewer than ten hidden layers can extract the required features needed to detect persons. Results reveal that the algorithm reaches a training accuracy of 96 % and a validation accuracy of around 97 %. In real-time detection, the system achieved an accuracy of 78 % in both illuminated and dark environments. The captured infrared images only capture thermal distribution with low image resolution and thus do not include face features or other privacy information. Due to its thermal imaging and edge computing capabilities, our system can also be extended to other application domains, especially in the post-pandemic era.

Publication Title

Building and Environment

Link to Full Text

Share

COinS