Measuring the Atmosphere: Environmental Sensing for Occupant Flows in Commercial Spaces
Naser Hossein Motlagh, Aygün Varol, Martha Arbayani Zaidan, Pak Lun Fung, Samu Varjonen, Petteri Nurmi and Sasu Tarkoma · IEEE Internet of Things Journal (2026)
Air-quality sensors already installed in a Helsinki shopping mall can tell how busy it is without cameras or phone tracking, using a model small enough to run on the sensor itself.
Abstract
Understanding visitor flows is important for commercial-space management, yet cameras, dedicated counters, and device-based probes create privacy, deployment, and coverage limitations. This paper presents an end-to-end environmental sensing system for scalable, privacy-preserving occupancy monitoring by repurposing existing building sensing infrastructure. We investigate environmental sensing as a low-cost alternative for estimating occupancy trends. We deployed 14 multivariable sensors at Tripla shopping mall in Helsinki, with each sensor operating for up to 79 days. We also co-located an automated people-counting camera with one of the sensors and used its measurements as ground-truth visitor counts. Our sensing study shows that individual variables, particularly CO2, cannot reliably capture rapid occupancy changes because of environmental latency, motivating the fusion of complementary sensing modalities. To efficiently exploit these heterogeneous signals, we develop a compact multi-modal attention-based inference pipeline suitable for resource-constrained edge devices. Under a temporally blocked split setting with disjoint multi-day training, validation, and test periods separated by gap days, the proposed system achieves 76.5% balanced accuracy in classifying Low, Medium, and High occupancy states, compared with 83.7% under a randomized split. Extreme Low-to-High misclassifications account for only 0.12% of test observations. The compact inference pipeline outperforms seven standard baselines under the blocked split setting while requiring only 5,813 parameters (approximately 23 kB), providing a favorable accuracy-model-size trade-off that supports inference on low-power edge hardware. Our ground-truth validation at one location, together with illustrative predictions for other mall zones, demonstrates the operational feasibility of the proposed sensing system in a single commercial building under realistic operating conditions. Overall, these results demonstrate the feasibility of an end-to-end environmental sensing system for computationally scalable and privacy-preserving visitor-count monitoring in commercial buildings.