People Counting and Desk Occupancy Sensors: How to Choose the Right Technology
People counting and occupancy sensors measure how many people use a space and where. Here is how PIR, radar, thermopile, ToF and AI sensors compare.
Occupancy data has become one of the most valuable inputs to a building. It tells estates teams which space is really used, lets the BMS heat, cool and ventilate only where people are, and gives retail and leisure operators accurate footfall. But "occupancy sensor" covers very different technologies, from a £50 PIR to a ceiling-mounted AI counter. This guide explains the options, where each fits, how to choose between PoE and LoRaWAN, and how to use the data in a BMS.
Occupancy detection vs people counting
Two related but different questions:
- Occupancy detection: is anyone here? A yes/no answer for a desk, cubicle, meeting room or zone.
- People counting: how many people are here, or have passed through this door? A number, often with direction (in/out).
| Need | Measure | Typical sensor |
|---|---|---|
| Is this desk in use? | Presence at a point | Desk sensor (PIR or thermopile under the desk) |
| Is this meeting room occupied? | Presence in a room | PIR, radar or thermopile ceiling sensor |
| How many people are in this room? | Count in an area | Thermopile or AI area counter |
| How many people entered the building or floor? | Count across a line | ToF or stereo vision counter over the entrance |
| Footfall into a shop | Count across a line | ToF, stereo vision or IR breakbeam |
Choosing the right question first avoids buying an expensive counter where a simple presence sensor would do, or the other way round.
Sensor technologies compared
PIR (passive infrared)
Detects movement of warm bodies. Cheap, low power and familiar from lighting controls.
- Strengths: low cost, long battery life, good for "is the room in use?"
- Weaknesses: detects motion, not presence, so a still person (reading, typing) can disappear; cannot count
- Examples: Milesight WS202 PIR and light sensor, WS203 motion, temperature and humidity sensor, VS340 desk and seat occupancy sensor
Radar (mmWave)
Detects tiny movements, including breathing, so it can sense people who are sitting still.
- Strengths: detects stationary presence, works in darkness, no images
- Weaknesses: more expensive than PIR; needs careful placement to avoid detecting movement outside the area
- Examples: Milesight VS370 radar human presence sensor
Thermopile (thermal array)
A low-resolution thermal "image" that sees heat sources as blobs, letting it count people in an area while preserving privacy.
- Strengths: counts stationary people, anonymous by design, battery options
- Weaknesses: lower accuracy in crowded or hot environments
- Examples: Milesight VS351 mini AI thermopile people counter
Time-of-flight (ToF)
Measures distance to objects with infrared light, building a depth map from above. AI algorithms identify and track people crossing a line.
- Strengths: high counting accuracy at doorways, works in any lighting, no identifiable images, can filter staff with badges
- Weaknesses: mounting height range and coverage width are limited per device
- Examples: Milesight VS133 AI ToF people counting sensor (PoE), VS135 ultra ToF people counter (LoRaWAN)
AI stereo vision
Two cameras compute depth (like human eyes) and AI tracks people in 3D. Images are processed on the device, and only counts leave it.
- Strengths: very high accuracy, wide coverage, high-ceiling mounting, multiple counting lines and zones, queue and dwell analytics
- Weaknesses: higher cost, usually PoE powered
- Examples: Milesight VS125 AI stereo vision people counter, VS126 ultra high-mount counter, and Xovis 3D sensors, which integrate with Niagara through the Tyrrell Products Xovis driver
AI workplace occupancy sensors
Ceiling sensors using on-device AI to count people and detect occupied desks across an area of an open-plan office.
IR breakbeam and specialist sensors
A beam across a doorway counts interruptions: simple and cheap for narrow single-file entrances (IR breakbeam people counter). Specialist variants include passage counters (VS350), storefront footfall sensors (VS361) and washroom cubicle occupancy sensors.
Summary table
| Technology | Detects still people | Counts | Typical accuracy | Privacy | Power |
|---|---|---|---|---|---|
| PIR | No | No | Presence only | Very high | Battery |
| Radar | Yes | Limited | Good for presence | Very high | Battery or mains |
| Thermopile | Yes | Yes (area) | Good | Very high | Battery or mains |
| ToF | Yes | Yes (line, area) | Very high at entrances | High (depth data only) | PoE or battery |
| AI stereo vision | Yes | Yes (line, zones) | Highest | High (on-device processing) | PoE |
| IR breakbeam | n/a | Yes (line) | Moderate; single file | Very high | Battery |
Desk and workplace occupancy sensors
Hybrid working has made desk occupancy one of the most searched smart building topics. Typical deployments:
- Under-desk sensors (PIR or thermopile) report each desk as occupied or free, for desk-booking apps and utilisation reporting.
- Ceiling AI sensors cover several desks per device, which is cheaper per desk in dense layouts and avoids fitting hardware to furniture.
- Meeting room sensors detect whether booked rooms are actually used, so no-show bookings can be released automatically.
Useful outputs include average and peak utilisation by floor, desk type and day of the week, which inform space consolidation decisions worth far more than the sensors themselves.
PoE vs LoRaWAN people counters
| PoE (Power over Ethernet) | LoRaWAN | |
|---|---|---|
| Power | From the network switch, no batteries | Battery or local power |
| Data | Real-time, high frequency | Periodic reports (e.g. every few minutes) |
| Installation | Needs a network cable to each sensor | Wireless; needs gateway coverage |
| IT involvement | Usually requires network ports and approval | Minimal |
| Best for | Entrances, high-accuracy counting, retail analytics | Retrofits, many rooms, areas without network cabling |
Many projects use both: PoE stereo vision or ToF counters at building and floor entrances, and LoRaWAN sensors in rooms and on desks. See What is LoRaWAN? and LoRaWAN range and planning.
Accuracy: what to expect and how to achieve it
High-end ToF and stereo vision counters can achieve very high counting accuracy at well-chosen entrances. Real-world accuracy depends heavily on installation:
- Mounting height within the manufacturer's range; too low narrows coverage, too high reduces accuracy
- Directly above the counting line, not at an angle
- Coverage of the full doorway width; wide entrances need several sensors or a high-mount model
- Avoiding reflections from glass and shiny floors (ToF) and strong backlighting (vision)
- Configuring counting lines and zones to match how people actually walk
- Excluding staff where needed with badges (for example the Milesight VA-series staff badges and lanyards)
Validate every counter after installation by manual counting over a busy period and adjust if necessary.
How many sensors do you need?
Plan from the question you want answered:
| Goal | Sensor plan |
|---|---|
| Building occupancy (how many people are in) | A line counter over every public entrance, including side doors and car park lobbies; total in minus out |
| Floor occupancy | Counters at stair and lift lobby entrances to each floor, or ceiling AI sensors across the floor |
| Meeting room use | One presence or area sensor per room; larger rooms may need two |
| Desk utilisation | One under-desk sensor per desk, or one ceiling AI sensor per group of desks within its coverage |
| Retail footfall | A counter over each customer entrance; wide entrances need several or a high-mount model |
| Washroom cleaning | Entrance counters per washroom plus cubicle occupancy sensors where needed |
The most common mistake in building-level counting is missing an entrance: a fire exit used as a shortcut or a staff door makes the in-minus-out total drift during the day. Most systems reset counts overnight to correct drift.
Commissioning checklist
- Mount at the specified height and directly above the counting line, using the correct bracket.
- Configure counting lines and zones in the device's web interface to match real walking paths.
- Set up staff exclusion if staff should not be counted.
- Check the network: PoE switch port and VLAN for PoE devices, or LoRaWAN join and signal quality for wireless ones.
- Validate accuracy with a manual count over at least one busy period and adjust lines or settings.
- Map outputs into the BMS with consistent names and tags (for example
floor,zone,occupancy,count). - Set reset times for daily totals and alarms for sensors that stop reporting.
- Record sensor positions and settings in the O&M documentation.
Privacy and GDPR
Occupancy and counting data is usually anonymous, but buildings in the UK must still consider UK GDPR:
- prefer technologies that do not capture identifiable images (PIR, radar, thermopile, ToF)
- where vision sensors are used, choose devices that process on-board and output only counts, with no image storage or streaming
- document the purpose (energy, space planning, safety), carry out a data protection impact assessment where appropriate, and tell occupants
- aggregate data for reporting rather than tracking individuals
Using occupancy data in the BMS
Occupancy data delivers most value when it drives control, not just dashboards. With Niagara 4:
- Heating and cooling setback: unoccupied zones drop to setback setpoints automatically, even during scheduled hours.
- Ventilation: people counts or occupancy reset ventilation demand alongside CO2 (see LoRaWAN IAQ and CO2 sensors).
- Lighting: lights off in empty areas.
- Cleaning and services: clean the washrooms and rooms that were actually used, using counts and smart washroom sensors.
- Reporting: utilisation by floor and building for estates decisions.
Integration routes include LoRaWAN (gateway or IONA-I), BACnet or Modbus from the sensor or gateway, MQTT, and dedicated Niagara drivers such as the Xovis driver and PointGrab integration. Watch the Xovis driver video in our Niagara 4 driver guides.
What does people counting cost?
The total cost of ownership has four parts:
- Hardware: from low-cost PIR and breakbeam sensors to premium stereo vision counters. Choose the cheapest technology that answers your question reliably.
- Installation: PoE sensors need a network cable and switch port per device; LoRaWAN sensors need gateway coverage but no cabling.
- Software and licences: some platforms (for example cloud analytics or vendor portals) are subscription based; integrating directly into Niagara keeps data in your BMS.
- Maintenance: battery replacements for wireless devices, validation checks after layout changes, and firmware updates.
The return usually comes from space decisions and energy: consolidating an underused floor or setting back unoccupied zones typically pays for the sensors many times over.
Applications by sector
Offices: desk and room utilisation, meeting room no-shows, occupancy-based HVAC.
Universities and schools: lecture theatre and library occupancy, space timetabling, ventilation control.
Retail and garden centres: footfall, conversion rates, queue monitoring, and heating and lighting matched to visitor numbers.
Leisure and venues: capacity management and ventilation by occupancy.
Healthcare: waiting room occupancy and patient flow.
Frequently asked questions
What is the difference between an occupancy sensor and a people counter?
An occupancy sensor answers whether a space is in use; a people counter measures how many people are in a space or have crossed a line.
How accurate are people counting sensors?
ToF and AI stereo vision counters can be very accurate at well-installed entrances. Accuracy depends on mounting height, position, doorway width and configuration, so validate after installation.
Do people counters record video?
Most modern counters process data on the device and output only counts. ToF, thermopile, radar and PIR sensors do not capture identifiable images at all.
Can a desk occupancy sensor detect someone sitting still?
PIR desk sensors can miss very still occupants; thermopile, radar and AI sensors detect stationary presence reliably.
Should I choose PoE or LoRaWAN people counters?
PoE suits entrances needing real-time, high-accuracy counting where network cabling exists. LoRaWAN suits retrofits and many rooms without network cabling.
What is the best sensor for counting people at a building entrance?
A ceiling-mounted ToF or AI stereo vision counter positioned directly above the entrance, such as the Milesight VS133 or VS125, or a Xovis sensor. Wide or high entrances need a high-mount model or several sensors.
Can occupancy sensors reduce energy costs?
Yes. Linking occupancy to the BMS lets heating, cooling, ventilation and lighting set back automatically in empty spaces, which is especially effective on hybrid-working estates.