The Privacy-First Sensor Fusion Stack: Why Camera-Free Occupancy Is Now the Fastest Path to Demand-Control HVAC

Researched by BEAST Library Curator | Verified against named industry sources (WebSearch-grounded) | Quality: 8.0/10

The Privacy-First Sensor Fusion Stack: Why Camera-Free Occupancy Is Now the Fastest Path to Demand-Control HVAC

BLUF: The bottleneck on occupancy-driven HVAC savings is no longer accuracy — thermal and 60GHz radar sensors are accurate enough. It's the legal and InfoSec approval queue. Camera-free fusion stacks (thermal + radar + CO₂) now clear privacy review in days instead of the weeks-to-months a camera-based system needs, and that approval speed is the real, under-counted line item in your ROI model. If you're scoping demand-control ventilation in the next 90 days, lead with the sensing modality your General Counsel will sign off on Friday — not the one with the prettiest dashboard.

The shift: fusion is mature, deployment is the constraint

For a decade the occupancy-sensing pitch was about accuracy — can the sensor count people in a room well enough to modulate airflow? That argument is largely over. The market has settled into three credible, named approaches, each with a different privacy posture:

VergeSense — AI camera-based sensors with computer vision and edge compute. Highest fidelity (exact headcount, desk-level), highest privacy scrutiny.
Density — 60GHz radar with depth sensing, anonymously counting people. No image ever captured.
Butlr — thermal sensors detecting only heat signatures, no visual or radio-frequency imaging at all.

Density's own framing is the tell on where the value has moved: "Because no PII ever enters the data pipeline, legal and InfoSec approvals that take weeks or months with camera-based systems can often be completed in days." That sentence is worth more to a facility GM than any accuracy spec sheet. A sensor that's 95% accurate but blocked in legal review for four months delivers zero kWh of savings; a sensor that's 88% accurate and live next week is already paying you back.

What "fusion" actually means at the BMS layer

Sensor fusion in a real building is not one magic device — it's combining complementary modalities so each covers the other's blind spot, then publishing a single occupancy signal into the BMS for control. The practical recipe most FMs can actually wire:

Modality	Best at	Blind spot	Privacy posture	Typical BMS path
PIR / motion	Cheap presence trigger	"Dies" when occupants sit still → false vacancy	Excellent (no identity)	Direct BACnet/Modbus
mmWave radar (60GHz)	Stationary presence + count, works in dark	Multipath in cluttered/glass rooms	Strong (no image, RF only)	IoT gateway → BACnet IP
Thermal (e.g. Butlr)	Count + movement, no imaging	Lower spatial resolution than camera	Strongest (heat only)	Wireless → gateway → BACnet
CO₂	Ventilation demand proxy (ASHRAE 62.1)	Lags real-time count by minutes	Excellent	Native BACnet/Modbus
Camera + CV (VergeSense)	Desk-level, highest fidelity	Slowest privacy approval	Highest scrutiny	Edge compute → API

The winning pattern for demand-control ventilation: pair a presence/count modality (radar or thermal) for real-time zone control with a CO₂ sensor for ASHRAE 62.1 compliance. CO₂ alone lags occupancy by several minutes — fine for ventilation setpoints, useless for fast lighting/VAV response. Radar or thermal gives you the fast edge; CO₂ gives you the code-defensible ventilation floor. That's fusion doing real work, not a buzzword.

Interoperability finally caught up

Two developments make this practical without rip-and-replace:

Matter 1.4 now enables interoperability across 800-plus certified devices, letting managers mix thermostats, lighting, and access badges across brands without middleware. This matters for retrofit: you're no longer locked into one vendor's sensor to talk to one vendor's controller.
LoRaWAN → BACnet bridging (e.g. MultiTech gateways) lets wireless sensors reach a legacy BMS without rewiring the building — the single biggest cost killer in older Class B/C stock. Wireless sensor → IoT gateway → BACnet IP or OPC-UA into Tridium Niagara, Siemens Desigo, JCI Metasys, Honeywell, or Schneider EcoStruxure.

The strategic read: BACnet/Modbus interoperability is shifting value capture from hardware margins to recurring energy-analytics contracts. The sensor is becoming a commodity; the fused signal and the control logic on top of it are the moat. (See our other Library analyses on the BaaS platform thesis for where that recurring-revenue layer lands.)

Edge AI is making fusion cheaper to run

A practical efficiency note for anyone worried about the data-handling overhead of multi-sensor fusion: edge processing now does the heavy lifting on-device rather than streaming raw data to the cloud. A recent multi-modal IoT node integrating 11 sensors (CO₂, VOCs, light, UV, pressure, temp, humidity, RGB, GNSS) on an ultra-low-power GAP9 SoC ran a YOLOv5 occupancy pipeline at the edge and demonstrated 42% energy savings over raw data streaming (arXiv:2507.14165). The takeaway for an FM: fusion no longer means a fat cloud bill and a bandwidth headache — the inference happens at the sensor.

APAC and data-center relevance

The same fusion logic is moving into AI data centers, where Taiwan's Delta unveiled (Data Center World 2026) a 3MW Liquid-to-Liquid Coolant Distribution Unit delivering up to 3,000 kW of cooling and 3,000 LPM flow, with native SNMP, Modbus TCP/IP, and BACnet for real-time monitoring. For APAC owners, the convergence is clear: cooling infrastructure is being shipped BMS-ready, and the occupancy/thermal sensor layer feeds the same control plane. Singapore's data-center and vertical-farm ventilation demand — where pressure-independent VAV outpaces established HVAC vendor coverage — is exactly the kind of high-density, high-stakes environment where fused real-time sensing earns its keep.

Here's what I'd do if this were my building

Start the privacy conversation before the procurement conversation. Take a thermal or radar (no-image) datasheet to your GC/InfoSec in week one. Get the "this is camera-free, heat/RF only, no PII" memo on file. That memo, not the sensor, is your critical path.
Pilot one floor, two modalities. Radar or thermal for fast zone presence + CO₂ for ventilation. Wire it into the existing BMS via a LoRaWAN→BACnet gateway — no rewiring. Budget a single floor, not the portfolio.
Instrument the M&V properly. Treat this as an IPMVP Option B or C exercise (retrofit isolation or whole-facility) from day one. Baseline ventilation/HVAC energy for 2–4 weeks pre-deployment, then measure demand-control savings against it. Don't accept a vendor's "up to X%" — measure your own.
Don't over-buy fidelity you can't use. Desk-level camera CV is overkill if all you're controlling is zone-level AHU/VAV. Match sensor fidelity to the smallest controllable HVAC zone — anything finer is paying for resolution the BMS can't act on.

The 15-year veteran's version: occupancy sensing stopped being a technology problem and became an approval-and-integration problem. The teams winning in 2026 aren't the ones with the most accurate sensor — they're the ones who picked the modality their lawyers approved fastest and wired it to the BMS they already own.

Have a question about this topic? Ask our CRE AI Agent →

Sources: Butlr — VergeSense vs Density vs Butlr; Butlr — PIR vs mmWave 2026; Delta CDU, Data Center World 2026; MultiTech LoRaWAN→BACnet; arXiv:2507.14165 multi-modal edge-AI node; PointGrab occupancy automation.