UK How people counting works
People counting is the automatic measurement of how many people enter and leave a space. A sensor watches each entrance, detects people as they pass and records the direction they travel. Entries minus exits gives live occupancy; entries over time give footfall, the visitor numbers that shops, stations and venues plan around.
Camera counters work in three steps. A detector finds each person in the frame, usually from directly overhead, where heads and shoulders are easy to separate. A tracker follows each person from frame to frame and gives them an ID. When a track crosses a virtual counting line, the system adds one to the in or out total. The tracking step is what lets a counter cope with groups, people who stop in the doorway and people who turn back.
Counts from several doors are combined in software, which shows live occupancy, hourly trends and alerts when a space nears its limit.
People counting sensors compared
Several technologies count people. They differ mainly in how well they separate people walking close together, and in what they capture.
| Sensor | How it counts | Accuracy in crowds | Privacy | Typical use |
|---|---|---|---|---|
| Infrared beam | Counts breaks in a beam across the door | Low: two people side by side count as one | No images | Small shops, simple footfall |
| Thermal sensor | Detects body heat from overhead | Medium | Low-resolution heat image only | Offices, meeting rooms |
| 2D AI camera | Detects and tracks people in video | Good, depends on angle and lighting | Video processed on the device or edge computer | Retail, venues, existing CCTV |
| 3D stereo or time-of-flight | Measures height to separate people, including children and groups | High, usually the most accurate option | Depth data, little identifiable detail | Stations, airports, large stores |
| Wi-Fi or Bluetooth | Counts phones broadcasting nearby | Estimate only, not every visitor carries a phone | Device identifiers, usually hashed | Dwell and repeat visits |
For accurate entrance counts, overhead 3D sensors and well-placed AI cameras lead. Beam counters remain common where an estimate is enough.
Why people counting is used
Counts feed decisions in safety, operations and planning. The same data often serves several teams.
-
//
Safety and occupancy limits. Stations, stadiums and venues keep crowds within safe limits and trigger alerts or entry control as a space fills.
-
//
Staffing. Shops and transport operators match staff and services to live and forecast demand.
-
//
Conversion and performance. Retailers divide transactions by visitors to measure how well a store turns visits into sales.
-
//
Building management. Offices adjust heating, ventilation and cleaning to actual occupancy, and plan space from real use.
-
//
Planning. Long-term footfall trends show the effect of timetable changes, events, layouts and opening hours.
What affects counting accuracy
Accuracy depends more on installation than on the brand of sensor. Mount counters overhead where possible, at the height the maker specifies, so people do not hide behind each other. Cover the full width of the entrance, and keep the counting line away from areas where people queue or linger. Lighting changes, low sun through glass doors, trolleys, pushchairs and children all need testing on site.
Good systems are checked against manual counts or video review after installation, and again after any change to the entrance. Camera-based counters are only as good as their tracking, which our object tracking FAQ explains. 3D counters rely on depth from two lenses, the principle behind a stereo camera.
Occupancy counting across several entrances
Live occupancy is simple in principle, entries minus exits, but small errors add up over a day. A counter that misses one exit in a hundred can show a building as busy long after it has emptied. Multi-door systems therefore reset occupancy at a known quiet time, such as closing, and some reconcile the total against a second source, for example a camera covering the whole floor.
Every entrance needs a counter, including staff doors and lifts from car parks, or the totals drift. The software then combines the doors into zones, so a station or venue can see occupancy per concourse, platform or hall, and set a different alert threshold for each.
Where the counting runs
Dedicated counters run the counting on the sensor and report totals over the network. Camera-based counting can also run on an edge computer that takes streams from several cameras, often the existing CCTV, and runs the detection and tracking models for every entrance at once. That keeps video on site, reports counts within a second or two and lets the same computer run other analytics, such as queue length or virtual fences.
Our NVIDIA Jetson PoE systems power and connect IP cameras from the same computer that runs the AI models. The station AI NVR case study includes occupancy counting at a metro station.
Quick answers on people counting
People counting: common questions
What is a people counter?
The device that does the counting. It is usually an overhead sensor or camera at an entrance that records people passing in each direction and reports totals over the network.
How accurate is people counting?
It depends on the sensor and installation. Overhead 3D and AI camera counters typically achieve high accuracy once tuned and checked against manual counts. Beam counters undercount groups.
Is people counting the same as footfall counting?
Largely, yes. Footfall is the retail term for visitor numbers. People counting also covers live occupancy and counting in non-retail spaces such as stations and offices.
Does people counting record faces?
Usually not. Overhead counters see the tops of heads, and most systems keep only counts. Camera systems still fall under data protection law in the UK and EU, so a privacy assessment is normally needed.
Can existing CCTV count people?
Sometimes. An edge AI computer can count from existing cameras if they see the entrance clearly. Angled CCTV views lose accuracy in crowds compared with an overhead counter.
We supply edge AI and PoE computers for camera-based counting as standard, with the operating system, AI runtime and camera drivers installed and tested before the system ships.