AI-Based Occupancy Detection for HVAC-Relevant Load Estimation in a University Building

Rufus  Dwamena; Wuxia Zhang; Paige Tien; John Calautit; Hao Sun

doi:10.65582/aifsc.2026.011

Authors

Rufus Dwamena Department of Architecture and Built Environment, University of Nottingham, United Kingdom.
Wuxia Zhang Department of Architecture and Built Environment, University of Nottingham, United Kingdom.
Paige Tien Department of Architecture and Built Environment, University of Nottingham, United Kingdom. https://orcid.org/0000-0003-0123-248X
John Calautit Department of Architecture and Built Environment, University of Nottingham, United Kingdom. https://orcid.org/0000-0001-7046-3308
Hao Sun Department of Architecture and Built Environment, University of Nottingham, United Kingdom; Global Centre for Clean Air Research (GCARE), University of Surrey, United Kingdom. https://orcid.org/0000-0002-9257-986X

DOI:

https://doi.org/10.65582/aifsc.2026.011

Keywords:

occupancy detection, HVAC control, computer vision, YOLOv7, EnergyPlus, seminar room

Abstract

Occupancy-aware HVAC control depends on realistic estimates of occupant-related heat gains, yet many buildings still operate on static schedules that assume sustained high occupancy. This study evaluated whether a vision-based people detector could generate a more representative occupancy schedule for a university seminar room. Two YOLOv7 models were trained on 500 annotated images, including five null images, and deployed to two simultaneous 60-minute recordings of room B5 in the Marmont Centre, University of Nottingham, from front-right and back-left viewpoints. Held-out validation favoured the extended Google Colab configuration over the Roboflow configuration, with mAP, precision and recall of 88.4%, 89.3% and 85.8% versus 86.9%, 84.1% and 80.9%, respectively. Against minute-by-minute manual counts, the best configuration - Google Colab with the front-right view - achieved a normalized count accuracy of 95.8%. Occupancy counts were translated into occupant heat-gain schedules using CIBSE benchmark data. The best dynamic schedule yielded a reconstructed mean occupant gain of 2.37 kW compared with 4.48 kW for a static seminar-room benchmark. Within the original EnergyPlus comparison framework, the best dynamic schedule corresponded to normalized winter-heating and summer-cooling indices of 73.3 and 32.4 relative to a static benchmark of 100, equivalent to reductions of 26.7% and 67.6%. The results show that camera placement and occlusion are as important as model choice and that vision-based occupancy detection can materially improve HVAC-relevant schedule realism in teaching spaces, provided that both counting accuracy and the direction of counting error are validated.

References

ASHRAE, 2020. ANSI/ASHRAE Standard 55-2020: Thermal Environmental Conditions for Human Occupancy. Atlanta, GA: American Society of Heating, Refrigerating and Air-Conditioning Engineers.

Bortolini, R, Rodrigues, R, Alavi, H, Vecchia, L F D and Forcada, N., 2022. Digital twins’ applications for building energy efficiency: A review. Energies, 15(19): 7002. DOI: doi.org/10.3390/en15197002.

Chen, Z, Jiang, C and Xie, L., 2018. Building occupancy estimation and detection: A review. Energy and Buildings, 169: 260-270. DOI: doi.org/10.1016/j.enbuild.2018.03.084.

CIBSE, 2021. Guide A: Environmental design (2015, updated 2021). Available at https://www.cibse.org/knowledge-research/knowledge-portal/guide-a-environmental-design-2015 [Last accessed 18 April 2026].

Naylor, S, Gillott, M and Lau, T., 2018. A review of occupant-centric building control strategies to reduce building energy use. Renewable and Sustainable Energy Reviews, 96: 1-10. DOI: doi.org/10.1016/j.rser.2018.07.019.

Pang, Z, O'Neill, Z, Chen, Y, Zhang, J, Cheng, H and Dong, B., 2023. Adopting occupancy-based HVAC controls in commercial building energy codes: Analysis of cost-effectiveness and decarbonization potential. Applied Energy, 349: 121594. DOI: doi.org/10.1016/j.apenergy.2023.121594.

Pincott, J, Tien, P W, Wei, S and Calautit, J K., 2022. Indoor fire detection utilizing computer vision-based strategies. Journal of Building Engineering, 61: 105154. DOI: doi.org/10.1016/j.jobe.2022.105154.

Rueda, L, Agbossou, K, Cardenas, A, Henao, N and Kelouwani, S, 2020. A comprehensive review of approaches to building occupancy detection. Building and Environment, 180: 106966. DOI: doi.org/10.1016/j.buildenv.2020.106966.

Tien, P W, Wei, S, Calautit, J K, Darkwa, J and Wood, C., 2020. A vision-based deep learning approach for the detection and prediction of occupancy heat emissions for demand-driven control solutions. Energy and Buildings, 226: 110386. DOI: doi.org/10.1016/j.enbuild.2020.110386.

Tien, P W, Wei, S and Calautit, J K., 2021a. A computer vision-based occupancy and equipment usage detection approach for reducing building energy demand. Energies, 14(1): 156. DOI: doi.org/10.3390/en14010156.

Tien, P W, Wei, S, Liu, T, Calautit, J, Darkwa, J and Wood, C., 2021b. A deep learning approach towards the detection and recognition of opening of windows for effective management of building ventilation heat losses and reducing space heating demand. Renewable Energy, 177: 603-625. DOI: doi.org/10.1016/j.renene.2021.05.155.

Tien, P W, Wei, S, Calautit, J K, Darkwa, J and Wood, C., 2021c. Occupancy heat gain detection and prediction using deep learning approach for reducing building energy demand. Journal of Sustainable Development of Energy, Water and Environment Systems, 9(3): 1-31. DOI: doi.org/10.13044/j.sdewes.d8.0378.

Tien, P W, Wei, S, Darkwa, J, Wood, C and Calautit, J K., 2022a. Machine learning and deep learning methods for enhancing building energy efficiency and indoor environmental quality - a review. Energy and AI, 10: 100198. DOI: doi.org/10.1016/j.egyai.2022.100198.

Tien, P W, Wei, S, Calautit, J K, Darkwa, J and Wood, C., 2022b. Real-time monitoring of occupancy activities and window opening within buildings using an integrated deep learning-based approach for reducing energy demand. Applied Energy, 308: 118336. DOI: doi.org/10.1016/j.apenergy.2021.118336.

UK Government, 2021. Approved Document F, Volume 2: Buildings other than dwellings, 2021 edition, 15 December 2021. Available at https://www.gov.uk/government/publications/ventilation-approved-document-f [Last accessed 18 April 2026].

UNEP and GlobalABC, 2025. Global Status Report for Buildings and Construction 2024/2025. Nairobi: United Nations Environment Programme.

U.S. Department of Energy, 2026. EnergyPlus, 31 March 2026. Available at https://energyplus.net/ [Last accessed 18 April 2026].

Wang, C-Y, Bochkovskiy, A and Liao, H-YM., 2023. YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Vancouver, Canada, 17-24 June 2023, pp. 7464-7475. DOI: doi.org/10.1109/CVPR52729.2023.00721.

Wang, F, Feng, Q, Chen, Z, Zhao, Q, Cheng, Z, Zou, J, Zhang, Y, Mai, J, Li, Y and Reeve, H., 2017. Predictive control of indoor environment using occupant number detected by video data and CO2 concentration. Energy and Buildings, 145: 155-162. DOI: doi.org/10.1016/j.enbuild.2017.04.014.

Wang, Z, Calautit, J, Wei, S, Tien, P W and Xia, L., 2022. Real-time building heat gains prediction and optimization of HVAC setpoint: An integrated framework. Journal of Building Engineering, 49: 104103. DOI: doi.org/10.1016/j.jobe.2022.104103.

Wang, Z, Calautit, J, Tien, P W, Wei, S, Zhang, W, Wu, Y and Xia, L., 2023b. An occupant-centric control strategy for indoor thermal comfort, air quality and energy management. Energy and Buildings, 285: 112899. DOI: doi.org/10.1016/j.enbuild.2023.112899.

Wei, S, Tien, P W, Calautit, J K, Wu, Y and Boukhanouf, R., 2020. Vision-based detection and prediction of equipment heat gains in commercial office buildings using a deep learning method. Applied Energy, 277: 115506. DOI: doi.org/10.1016/j.apenergy.2020.115506.

Wei, S, Tien, P W, Wu, Y and Calautit, J K., 2022a. A coupled deep learning-based internal heat gains detection and prediction method for energy-efficient office building operation. Journal of Building Engineering, 47: 103778. DOI: doi.org/10.1016/j.jobe.2021.103778.

Wei, S, Tien, P W, Chow, T W, Wu, Y and Calautit, J K., 2022b. Deep learning and computer vision based occupancy CO2 level prediction for demand-controlled ventilation (DCV). Journal of Building Engineering, 56: 104715. DOI: doi.org/10.1016/j.jobe.2022.104715.

Zhang, W, Calautit, J, Tien, P W, Wu, Y and Wei, S, 2024. Deep learning models for vision-based occupancy detection in high occupancy buildings. Journal of Building Engineering, 98: 111355. DOI: doi.org/10.1016/j.jobe.2024.111355.