Precision in the air: A lightweight, cost-effective, CFD-assisted method for occupant localization and demand-based airflow control in crowded indoor spaces

Improving occupant comfort while reducing energy use is a key challenge in buildings. In this context, zonal demand-controlled ventilation with demand-based airflow offers an energy-efficient alternative to conventional heating, ventilation, and air conditioning systems. However, achieving effective airflow distribution requires knowing occupant positions. To address this, this study proposes a lightweight, low-cost, and localized occupancy positioning method to support occupant-aware airflow control in indoor environments. For this, an occupancy detection model based on You Only Look Once version 12 nano (YOLO12n) was trained for dense environments and deployed on an NVIDIA Jetson using RGB and stereo cameras. Camera calibration ensured accurate localization, with reprojection errors (0.1103 pixels for RGB; 0.0932 pixels for stereo) well below the 0.2 pixels benchmark for precision. Localization experiments across meeting, office, and seminar spaces showed the stereo camera achieved higher accuracy (3.43 % and 2.21 % relative error along the X and Z axes), while the RGB camera performed acceptably at distances below 4 m (5.98 % and 2.87 % error) as a cost-effective solution. Both demonstrated robust performance under occupant movement, occlusions, and varying lighting conditions. To assess thermal comfort, computational fluid dynamics simulations were conducted to identify optimal fan speeds and diffuser angles that ensure air mixing and reduce draft discomfort. Best- and least-performing configurations, adjusted based on occupant localization results, were validated through a survey involving nine participants in the same room used for the localization experiments. Results showed that demand-based airflow reduced negative thermal sensation feedback by 29.6 %. This study demonstrates that vision-based positioning enables effective airflow control.