A station concourse or an outdoor venue is the hardest kind of space to measure, and the one where getting it wrong is most dangerous. When tens of thousands of people move through a transit hall or a festival arena, a headcount at the gate tells you almost nothing that matters. What matters is density, flow and dwell, right now, per zone. This is a deep dive on concourse monitoring and outdoor venue occupancy analytics with edge computer vision, and why counting a crowd is a very different problem from counting an office.
01 · WHAT CONCOURSE MONITORING ACTUALLY MEASURESWhat concourse monitoring actually measures
Concourse monitoring is not a single number. A safe, well-run concourse is described by four measurements, read together and read live:
- Density, in persons per square metre, is the safety-critical one. It is the variable crowd scientists watch because it predicts crushing risk before anything visibly goes wrong.
- Flow, in people per minute and by direction, tells you whether a space is filling faster than it can clear.
- Dwell, how long people linger, separates a thoroughfare from a bottleneck.
- Queue length at gates, gatelines and stairs is where flow and density turn into a decision.
The thresholds are not arbitrary. In his foundational work on pedestrian planning, John Fruin defined a Level of Service scale from A to F, and reported that at roughly seven persons per square metre a crowd behaves almost like a fluid, with shock waves propagating through it and individual control of movement lost. The UK's Guide to Safety at Sports Grounds, known as the Green Guide, is built around calculating safe capacity, and practitioners treat densities above six persons per square metre as congested and unstable, the point where injuries begin.
Read those numbers against a live density map and concourse monitoring stops being CCTV that someone watches and becomes an early-warning system that watches itself. For the operational side of getting a crowd out safely, we went through the arithmetic in egress under twelve minutes and pacing venue exits to train platform capacity.
02 · WHY OUTDOOR VENUES BREAK TRADITIONAL SENSORSWhy outdoor venues break traditional sensors
Everything that makes indoor people counting tractable disappears outdoors. There is no doorway to count through, no fixed gateline, no ceiling to mount to. The crowd is transient, the boundary is a fence line across a field, and the weather does not cooperate. Outdoor venue occupancy analytics is a different discipline, and most sensing approaches quietly fail at it.
- Turnstiles and badge data count entries, not where people actually are twenty minutes later.
- Wi-Fi and Bluetooth counting need dense access-point coverage that outdoor sites rarely have, miss anyone without a phone, and break under address randomisation.
- Cloud video is accurate but does not scale. Streaming hundreds of camera feeds off a greenfield site to a data centre is a bandwidth and a privacy problem at the same time, and it puts identifiable footage of a crowd into storage.
Measuring an outdoor venue means putting the intelligence at the edge, distributed across the site, producing anonymous numbers rather than shipping video back to be watched.
03 · EDGE COMPUTER VISION: A PLATFORM FOR VIDEO AND SENSOR ANALYTICSEdge computer vision: a platform for video and sensor analytics
This is the part that makes crowd-scale monitoring possible. Edge computer vision runs the model on the device that holds the camera. Each frame is inferenced locally, a count and a density estimate come out, and the frame is discarded in the same tick. What leaves the device is roughly 200 bytes per second of anonymous JSON, not a video stream. No footage is stored, none is transmitted, and there is no face model anywhere in the pipeline.
Two properties fall out of that design, and both matter at venue scale:
- It scales. Because each node ships counts instead of video, you can put hundreds of sensors across a concourse or a festival site without a data-centre-sized uplink. The bandwidth of the whole deployment is a rounding error.
- It fuses inputs. An edge AI platform for video and sensor analytics does not have to stop at cameras. The same edge node can combine the vision count with other signals on site, so the occupancy layer is one coherent number per zone rather than a pile of disagreeing feeds.
That is what our on-device edge AI and the camera-free occupancy sensor are built to do. The sensor has the optical accuracy of a camera and the privacy profile of a presence detector, because the image never leaves the room, or in this case, the field.
04 · DENSITY, FLOW, AND DWELL TIME IN PRACTICEDensity, flow, and dwell time in practice
The same three measurements drive different decisions depending on the space.
Density drives safety. On a concourse it is the number that trips an alert before a crush forms, giving a control room minutes instead of seconds to open a gate or hold ingress.
Flow drives operations. Direction-aware counts expose the bottleneck a static plan never predicted, the stair everyone takes and the exit nobody walks to.
Dwell drives revenue. This is where the retail world has done the maths most rigorously. Dwell time is one of the strongest predictors of spend: grocery shoppers average around 22 to 25 minutes in store while apparel shoppers sit closer to 8 to 12, and analyses of major chains put average visit dwell at 37 minutes in Costco against roughly 29 in Target. The correlation is direct: shoppers who stay a few extra minutes convert far more often than pass-through browsers. We pulled that thread in dwell time and the truth about merchandising, and the same dwell signal feeds live occupancy analytics across every space we work in.
05 · GETTING THE DATA OUT: THE OCCUPANCY APIGetting the data out: the occupancy API
A count that stays inside the sensor is worthless. The reason people search for an occupant or occupancy API is that the number only earns its keep when it reaches the systems that act on it: a control-room dashboard, a building management system, an alerting stack, a data warehouse.
So the output is designed to be consumed, not admired. Anonymous counts and threshold events are exposed over a REST endpoint, an MQTT topic, and webhooks over HTTPS, which is the same plumbing that lets the platform push into a building management system, Slack, Teams or Snowflake. See the full list of integrations for the connectors, wire threshold alerts for the safety-critical density lines, and, because the data is queryable, you can ask the building a question in plain English instead of writing a report.
06 · WHERE IT RUNSWhere it runs
The method is one thing; the rhythm changes with the venue.
- Stations and transport interchanges, where concourse density and platform capacity have to be balanced in real time. That is the world of transport.
- Stadiums and arenas, where ingress, egress and concourse safety are regulated and audited. See events and stadiums.
- Festivals, civic squares and outdoor events, run by councils and promoters who own the duty of care. See cities and councils.
- Retail and transit retail, where the same counts turn into dwell and conversion. See retail.
07 · PRIVACY AT CROWD SCALEPrivacy at crowd scale
Counting a crowd must never mean identifying the people in it. This is the line that separates crowd analytics from surveillance, and it is easiest to cross precisely where the crowd is largest. An edge-inference design holds the line by construction: there is no biometric template, no demographic classifier, and no stored image to be re-identified later. The output is anonymous by architecture, which is what makes it defensible under the Australian Privacy Act and GDPR rather than merely compliant on paper.
We built a whole event around exactly this point, monitoring a civic festival without facial recognition, in civic festival safety without facial recognition, and the underlying guarantee is documented in privacy by design.
08 · SOURCES AND FURTHER READINGSources and further reading
- John Fruin, Levels of Service, via Prof. G. Keith Still, on crowd density and the A to F scale.
- Guide to Safety at Sports Grounds (Green Guide) and calculating safe capacity, Sports Grounds Safety Authority.
- Retail dwell time metrics, Retail Sensing.
- Consumer time spent in store, eMarketer, on dwell benchmarks across major chains.
Concourse monitoring and outdoor venue occupancy analytics come down to one idea: put the intelligence at the edge, measure density, flow and dwell anonymously, and get the number to the people who have to act on it before the crowd forces the decision. If you run a station, a stadium, a festival or a shopping centre and you want to see that working in your own space, without a single frame of video leaving it, book a demo and we will walk it with you.