Home / Cities and public spaces / Crowd monitoring

Crowd monitoring that warns staff before an area gets too full.

A crowd at a station or event gate builds in minutes, and staff often see it too late. ZoneIntel AI runs crowd monitoring on your CCTV and alerts the duty manager while staff still have time to open gates.

Crowd build-up18:07
Crowd monitoring alert: concourse camera from above with people counted inside a marked area

Concourse area C2 holds 16 people, about 4 per square metre. The station set its first level at 3.

Where
Concourse camera C2, east entrance
Sent to
Station duty manager, by Microsoft Teams and the dashboard
If nobody responds
The operations supervisor is alerted after 5 minutes above the level, then the city operations centre.
Example alert with made-up names. Simulation.

Walking and standing people are counted apart

A standing group at a gate line builds a crowd faster than people walking through. ZoneIntel counts both and shows staff where groups have stopped.

Simulation. Station footage from Pexels, under the Pexels licence.

Levels set by the venue

Each area has a size and a safe level. ZoneIntel reports people per square metre and the trend over the last few minutes. Alerts follow the levels below.

Level in the areaWhat staff seeTypical response
Under the first levelCount and trend on the dashboardNo action
First level reachedAlert to the duty manager with the clipOpen more gates, deploy staff
Safe level reachedAlert escalated to the operations supervisorHold entry, use other routes
Back under the first levelCase closed with the photoNormal operation

Results

ResultPilot measure
Build-up seen early enough to actMinutes between the first alert and the safe level
Staff sent to the right gateAlerts answered within the set time
A record for every event dayPeak counts per area, with clips

Staff open gates and change routes before people are pressed together. After each event, the venue has a record for its safety review.

  • For the 2025 Hajj season, SDAIA’s Sawaher platform linked more than 5,000 cameras across more than 80 sites.

    Source: Arab News

Closed with photo evidence

A crowd case closes when the count in the area is back under the first level, with a photo from the same camera.

Concourse camera with people counted inside area C2
18:07, first level passed
The same concourse camera with fewer people in area C2
Back under the first level
Case C2-0907Concourse camera C2, east entrance
  1. 18:07Area C2 reaches about 4 people per square metre. The duty manager is alerted.
  2. 18:09Staff open the second gate and send people down the west stairs.
  3. 18:15The count falls as the next train leaves.
  4. 18:21Back under the first level, and the case closes with the photo.

Questions about crowd monitoring

How does crowd monitoring work on CCTV?

Crowd monitoring counts the people inside a marked area and divides by its size. That gives people per square metre. ZoneIntel compares the figure with the venue’s level and alerts staff before it is reached.

Does ZoneIntel AI identify people in a crowd?

Crowd density analytics count people and never identify them. They keep no face templates or identity records, and shared clips show blurred faces. This helps the city's privacy review for each public space.

What density level should trigger an alert?

The venue sets its own levels with its safety adviser. Many use a first alert well below the planned maximum and a second at the maximum. That gives staff time to open gates or hold entry.

Can it count people walking through a gate?

ZoneIntel counts people crossing a line drawn on the camera, in each direction. Gates and corridors report flow per minute next to the density in the areas around them.

Does it work for temporary events?

Temporary events use cameras added for the event and a ZoneIntel server on site. Areas, levels and alert routes are set up before the gates open.

Crowd count from your own footage

Share footage from one busy day through a secure link. We send back counts per area, peaks and the alerts ZoneIntel AI would have raised.