Why Occupancy Sensing Is Harder Than It Looks
Every occupancy technology fails in a characteristic way, and the failures are not random. How to combine sensors so the mistakes cancel instead of compounding.
Occupancy detection sounds like a solved problem and is one of the most persistent sources of complaints in building automation. The reason is that every available technology answers a slightly different question from the one the building actually asks, and the gaps between those questions are where the annoyance lives.
Passive infrared, the most common sensor, does not detect people. It detects changes in received infrared across its field of view, which correlates with a warm body moving. A person typing quietly is close to invisible to it. This is why lights go out on people sitting still in offices, and why the fix so many buildings adopt — extending the timeout — trades one failure for wasted energy across every genuinely empty room.
CO2 measures a consequence of occupancy rather than occupancy, and consequences lag. Concentration rises over ten or twenty minutes as people breathe and falls slowly afterwards, which makes it excellent evidence that a room has been occupied for a while and poor evidence about the last few minutes. A control loop driven by CO2 alone always arrives late in both directions.
Cameras and radar answer the question most directly and carry a different cost. A camera can count people accurately, and it also creates a device that watches a room — a privacy conversation with employees, a data protection assessment, and an attack surface. Millimetre-wave radar detects presence through breathing and small motion without producing an image, which sidesteps most of that, at higher unit cost and with its own tuning difficulties around adjacent spaces and through partitions.
The practical approach is to fuse on the basis of complementary failure modes rather than to seek a better single sensor. Motion responds immediately and under-reports stillness. CO2 responds slowly and cannot be fooled by stillness. Combining them means the fast signal triggers and the slow one sustains, so an occupied room stays conditioned and an empty one does not linger. Door counting, badge readers or calendar data can be added as further weak evidence when available.
The last thing worth designing deliberately is what happens when the system is wrong, because it will be. Failing toward comfort — staying conditioned a little longer than strictly needed — costs energy. Failing toward efficiency costs occupant trust, and a building whose occupants have learned to distrust the automation gets overridden manually, which loses far more energy than the conservative setting ever would.