Crowd density at events: when does it get too crowded?

In front of the stage, at the entrance gates, at the exits once the last song has ended: almost every event has places where people bunch up. Whether that is merely busy or already dangerous is hard to judge by eye – particularly from a distance, at night, or on a camera image with no sense of scale.

The question that matters is therefore not “how many people are there?” but “how much room does each person have?”. That is exactly what crowd density describes.

What crowd density means

Crowd density is the number of people standing or walking on one square metre. Its inverse is the space per person: at 2 people per m², everyone has roughly half a square metre; at 4 people per m², only a quarter.

A headcount alone says little: 5,000 people on a large site pose no problem, yet 400 of them on a narrow access route can. Density always refers to a specific place.

The basics: Fruin’s Level of Service

The best-known classification comes from John J. Fruin. In Pedestrian Planning and Design (1971) he defined six Levels of Service (LOS), from A to F. For people standing or queuing, the approximate bands are:

Level	Space per person	Density	Conditions
A	over 1.2 m²	below 0.83 p/m²	free movement
B	approx. 0.93–1.2 m²	0.83–1.08 p/m²	comfortable standing, movement partly restricted
C	approx. 0.65–0.93 m²	1.08–1.54 p/m²	standing comfort reduced, limited circulation
D	approx. 0.28–0.65 m²	1.54–3.57 p/m²	no movement without contact
E	approx. 0.19–0.28 m²	3.57–5.26 p/m²	unavoidable body contact
F	under approx. 0.19 m²	over 5.26 p/m²	no independent movement possible

The key lesson is how narrow the gap is between “comfortable” and “critical”. Moving from level C to level E takes only around three additional people per square metre.

G. Keith Still’s rule of thumb

The British crowd scientist G. Keith Still (2000, 2014) has given practitioners a memorable rule of thumb:

• from about 2 people per m², movement is restricted;
• from about 4 people per m², conditions become dangerous;
• from about 6 people per m², the crowd is crushing.

The HSE’s Event Safety Guide (1999), widely known as the “Purple Guide”, points in the same direction, recommending a maximum density of 2 people per m² for open event areas.

Not every location tolerates the same density

A single limit for the whole site falls short. On walkways and at entrances people need to keep moving, so congestion sets in far earlier than in a standing audience; seating has its own standards again. Each observed location therefore deserves its own warning profile.

Measuring density in practice

To monitor density you need two figures: the number of people in an area, and the size of that area in square metres. Both are harder to obtain than they sound.

Counting people: heads rather than bodies

At low density, whole bodies are easy to detect in a camera image. As the crowd tightens, bodies increasingly hide one another – but seen from above, heads usually remain visible. Head detection is therefore the right method for dense crowds. The CrowdHuman dataset (Shao et al., 2018) was built specifically for detecting people in crowds and includes head annotations for this purpose.

Camera placement matters a great deal. A steep viewing angle of roughly 30 to 60 degrees from the vertical makes head detection easier, and good lighting or infrared illumination helps after dark.

Determining area: measuring the ground

A camera image has no scale. A square metre in the foreground covers far more of the picture than one 30 metres away. To turn a head count into a density in people per m², the observed ground has to be measured: you mark four points on the ground in the camera image whose positions in metres are known – the corners of a measured rectangle, for example – and enter the camera’s mounting height.

From this, the area of each observed zone can be calculated in square metres. It also corrects an error that is easily overlooked: a head at 1.6 metres appears considerably further back in the image than the same person’s feet. With the camera mounted at 5 metres, the shift amounts to around half the distance. Without this correction, people near the edge of a zone would be assigned to the wrong side systematically.

Without measurement: occupancy instead of density

Without ground measurement, true density cannot be reported. The fallback is a maximum headcount per zone, with occupancy assessed as a percentage of it – for instance with levels at 50, 70 and 90 per cent. Workable, but only as good as the estimated maximum.

Steadying the readings

Individual frames fluctuate as people are briefly hidden or counted twice. To stop warnings flickering on and off, raw values should be smoothed, for example as a rolling median over a few seconds, combined with hysteresis: a level only drops back once the value is clearly below the threshold.

From single camera to site map

Each camera sees only its own section, but event control needs the whole site. A heat map combines several cameras’ readings on one map. Between measuring points you can only estimate, and areas no camera covers should be marked “not observed” rather than filled with invented values.

How MOVETOS OPS captures crowd density

MOVETOS OPS applies these principles within the platform:

• Head counting in the camera image: in dense crowds the count relies on detected heads; with few people it uses full-body detection.
• Measuring the ground: when setting up a camera, you mark four ground points and enter the mounting height. The platform then calculates the area of each observed zone, projects head positions down to ground level and shows density directly in people per m² and as a Fruin level. Without measurement, it assesses occupancy against a defined maximum.
• Warning profiles per location: every observed location has a profile. The presets are “Walkways and entrances” (busy from 1.2, very busy from 2.0 people per m²), “Standing audience” (2.5 / 3.5) and “Seating” (1.2 / 1.8); custom values are possible.
• Four plain-language levels: on the site map, conditions are shown as “clear”, “lively”, “busy” and “very busy”, always alongside the figure. Readings are smoothed over three seconds, and a level only drops once the value is 15 per cent below the threshold.
• Density field across the event area: measured zones are placed on the map using camera position, viewing direction and ground measurement. Values between readings are estimated; areas well beyond any camera’s range are hatched grey and marked “not observed”.
• Warnings as suggestions: when an area becomes busy, the warning appears as a sentence at the top of the situation view and, if desired, as a notification on mobile phones. A suitable announcement is suggested but never played automatically. The system suggests, people decide. Each warning is recorded in the operations log with a timestamp and a still image.

After the event, the analysis shows for each location when it was fullest and how long it was “busy” or “very busy”, with the thresholds drawn as lines on the timeline.

Checklist for your event

1. Identify critical locations: entrances, bottlenecks, the stage area, exits – where could density build up?
2. Choose a standard for each location: walkways and entrances need lower thresholds than a standing audience.
3. Mount cameras appropriately: high and at a steep angle, with a clear view of heads.
4. Measure the ground: only then does a head count become a reliable density in people per m².
5. Prepare your procedures: who responds at “busy”, who at “very busy”? Which announcement fits, which entrances can be diverted?
6. Know your blind spots: be aware of which areas no camera covers, and plan staff for them.

Conclusion

You cannot tell when a crowd becomes too dense from the attendance figure – only from the space each person has, location by location. Fruin’s Level of Service scale and Still’s rule of thumb provide sound reference points. Cameras can measure density reliably once heads are counted and the ground area has been measured. Deciding what happens next remains the job of event control.