An event crowd analytics system counts people entering and leaving defined spaces using overhead cameras, thermal sensors or LiDAR, and reports live occupancy against declared capacity, crowd density, flow rate through chokepoints and queue length. Counting is anonymous — no faces are identified — and thresholds raise alerts to operations before a space becomes unsafe rather than after.
Built for the person who signs the event safety plan first, and for the marketing report second. Those are different requirements and the safety one sets the specification.
- Live occupancy against declared capacity
- Threshold alerts before a space becomes unsafe
- Anonymous counting — no facial recognition
- Retained evidence for the post-event safety review
The number nobody can actually produce
Ask a venue how many people are inside hall 3 right now and the honest answer is usually an estimate. Tickets sold is not occupancy — it ignores no-shows, re-entry and people who left. A clicker at the door drifts within an hour and counts nobody leaving. CCTV shows a busy room but not a number. So the capacity figure in the safety plan is managed by stewards' judgement, which works until the evening a headline act, a prayer time and a hall closure coincide.
The second gap is timing. Crowd problems are visible on camera about two minutes after they become difficult to fix — by then the chokepoint is already full and the only remaining options are stopping entry or opening a route that was not planned. What operations needs is the density trend fifteen minutes earlier, while stewards can still be moved. The third gap is evidence: after an incident or a near-miss, there is often no record of what the occupancy actually was, which makes the review a matter of recollection.
- Tickets sold is not occupancy; it ignores no-shows, exits and re-entry.
- Manual counting drifts within an hour and rarely counts people leaving.
- By the time congestion is visible on CCTV it is already hard to relieve.
- Chokepoint flow rate, not floor area, is what limits safe throughput.
- Without a retained record, a post-incident review is recollection, not evidence.
Solution overview
Swedish Technology deploys anonymous counting at every boundary that matters — entrances, hall thresholds, stair and escalator heads, theatre doors — and maintains a live occupancy figure per space as a running balance of entries minus exits. Overhead sensors are used rather than face-height cameras: an overhead view counts heads reliably in a crowd, is not confused by people standing in front of one another, and cannot identify anyone.
Occupancy is reported against the declared capacity for that space, with amber and red thresholds agreed with the safety officer and alerts routed to the people who can act. Alongside occupancy, the system measures density in defined zones, flow rate through chokepoints and queue length at entry points — the three measures that indicate a developing problem before the count itself becomes alarming. All inference runs on site, and only counts leave the venue network.
How the solution works
- 1Define spaces and capacities Each monitored space, its declared capacity and its amber and red thresholds are agreed with the safety officer and written into the configuration — this is a safety document, not a settings screen.
- 2Place sensors at boundaries Overhead sensors at every entrance and exit of each space, because occupancy is a balance and an unmonitored exit makes the count drift upward all day.
- 3Count anonymously Edge inference detects and tracks head-shaped objects crossing a line, producing directional counts. No faces, no identities, no images retained.
- 4Maintain the balance Entries minus exits per space, with periodic reconciliation against a known-empty state — typically overnight — so small errors cannot accumulate across days.
- 5Measure density and flow Density in persons per square metre for defined zones, flow rate through chokepoints and queue length at entries, sampled continuously.
- 6Alert and record Threshold breaches alert operations by dashboard, SMS or radio integration, and every reading is retained for the post-event review.
Key capabilities
Reference architecture
Inference at the edge, counts to the centre. Video never leaves the sensor, which is what makes the privacy position simple and the bandwidth requirement trivial.
Deployment options: On-premise by default. Inference runs on the sensor or a local edge server and only aggregate counts are stored, so the system continues to alert during an internet outage — which is precisely when a venue can least afford blind spots.
Hardware options
Sensor choice follows the environment and the accuracy the safety case needs. Overhead mounting is the constant; the sensing technology is not.
| Device | Where it is used | Selection notes |
|---|---|---|
| Overhead stereo-vision counter | Doors, hall thresholds, gate lines | The default for controlled indoor entrances. Typically 95–98% accurate at normal flow; degrades as people pass shoulder to shoulder. |
| Thermal counting sensor | Entrances with difficult or variable lighting | Unaffected by darkness or glare and inherently anonymous, at lower resolution. Suits outdoor gates and dim venues. |
| Overhead LiDAR | Wide openings and dense flow | Highest accuracy in crowds and across wide spans, at the highest cost. Used where a single opening carries a large share of total flow. |
| Existing CCTV camera | Open areas where zone density matters more than a line count | Reuses installed cameras for density estimation. Accuracy depends on angle and height; a face-height camera cannot count a dense crowd reliably. |
| Edge inference server | Venue rack room | Where analytics run centrally on multiple camera feeds rather than on-sensor. Sized by stream count, and kept on the venue network. |
Swedish Technology supplies and integrates counting equipment from established manufacturers; sensor selection follows the site survey, the required accuracy and current availability.
AI capabilities
The models here are deliberately narrow: detect a person, track them across a line, and forecast the next twenty minutes.
- Overhead person detection — Detects and tracks head-and-shoulder shapes from above, which stays reliable in crowds where a face-height view fails because people occlude one another.
- Density estimation — Estimates persons per square metre in an open zone from an angled camera, giving a usable density figure where line counting is not possible.
- Occupancy forecasting — Projects occupancy fifteen to thirty minutes ahead from current arrival and departure rates, which is the window in which stewarding can still be changed.
- Anomaly and stall detection — Flags flow that stops unexpectedly at a chokepoint or a counting line whose rate diverges from its neighbours — often the first sign of an obstruction or a failed sensor.
Integrations
Occupancy data is most useful where it changes something automatically. These integrations can be designed within project scope.
| System | Integration point & data exchanged | Direction |
|---|---|---|
| CCTV / VMS | A threshold alert can call up the relevant camera view in the control room, so the operator sees the space at the moment the alert fires. → RAQEEB – AI Surveillance | bi-directional |
| Access control and turnstiles | Entry can be paused or metered automatically when a space reaches its red threshold, and released when occupancy falls back. | bi-directional |
| Digital signage | Live wait times and 'hall full — please use hall 2' messaging published to entrance screens, which redistributes crowds without a steward. → Smart Queue Management for Events | outbound |
| RFID visitor tracking | Credential reads and anonymous counts at the same doorway validate each other — two independent measurements are far more defensible than one. → RFID Visitor Tracking for Exhibitions | bi-directional |
| BMS / HVAC | Occupancy can drive ventilation and cooling set-points, which matters in a Gulf venue where a full hall changes the thermal load quickly. → Facility Management | outbound |
| Incident management / Teams | Alerts and acknowledgements logged into the venue's incident system so the response is recorded, not just the alert. | outbound |
The integrations above are designed and implemented within project scope using vendor APIs, webhooks or standard connectors. They do not imply partnership, certification or endorsement by the system owner unless stated on that vendor's official pages.
Dashboards & analytics
- Control room (live) — Occupancy per space against capacity with amber and red state, density by zone, chokepoint flow rates and open alerts awaiting acknowledgement.
- Entry operations — Arrival rate per gate, queue length and estimated wait, and lanes open versus lanes needed for the current rate.
- Safety review — Time-series occupancy and density for any window, alert and acknowledgement history, and the evidence pack for a post-event debrief.
- Commercial — Footfall by zone and hour, dwell distribution and repeat-visit patterns — reported after the safety views, because it is the secondary purpose.
Security & deployment
All inference runs on the sensor or on a local server inside the venue network, and only counts leave it — there is no video stream to the cloud and no image retained. That keeps the system operating during an internet outage, which is when a venue can least afford to lose its occupancy figure. Access is role-based, with the safety officer able to change thresholds and operators able only to acknowledge alerts, and every threshold change is logged because it is a change to a safety control.
Data privacy
Counting is anonymous by construction, not by policy. Overhead sensors see the top of a head and shoulders; thermal sensors see a heat signature. No facial image is captured, no biometric template is created, no identity is resolved and no video is retained — the stored record is a time series of numbers.
Where existing CCTV is reused for density estimation, that footage remains under the venue's existing CCTV regime and retention, and the analytics layer takes counts from it without adding a new retained copy. Under UAE Federal Decree-Law No. 45 of 2021 the venue is the controller; because the analytics output contains no personal data, the compliance burden sits with the underlying CCTV rather than with the counting system.
Industry use cases
UAE & GCC considerations
Crowd safety for public events in the UAE is subject to permitting and to civil defence requirements, and Dubai venues additionally operate within SIRA's framework for security systems and personnel. In practice this means an occupancy system is most useful when it produces evidence in the form the authority expects — declared capacity per space, a measured figure against it, and a retained record — rather than a marketing heat map. Final compliance positions should be confirmed with the relevant authority and the client's own advisers.
The regional operating pattern matters too. Very large arrival spikes are normal here — a hall can take a third of its day's attendance in ninety minutes — and prayer times, evening peaks and the Ramadan calendar shift those spikes in ways a generic forecast will not anticipate. Outdoor and semi-outdoor gates also need sensors rated for high ambient temperature and dust, and thermal counting performs better than vision at an unlit outdoor gate at night.
Implementation approach
- 1Safety scoping Spaces, declared capacities, thresholds and who is alerted — agreed with the safety officer, because these are safety controls rather than configuration preferences.
- 2Site survey Every entrance and exit of each monitored space identified, with mounting heights, opening widths, lighting and power assessed. An unmonitored exit invalidates the occupancy balance.
- 3Sensor selection and install Technology chosen per opening — stereo vision, thermal or LiDAR — installed overhead and aligned to the counting line.
- 4Calibration and accuracy test Controlled walk-throughs with a known count establish real accuracy per sensor, under load rather than empty.
- 5Alerting and integration Thresholds, escalation paths and integrations to VMS, signage and access control built and tested with the control room team.
- 6Live event and review Monitored through the first event with an engineer present, then a review comparing forecast to actual and tuning thresholds from real data.
Why Swedish Technology
- We specify from the safety case first, so the system produces the number the safety officer is accountable for rather than a marketing heat map.
- Every exit is counted, because an occupancy balance with an unmonitored exit drifts upward all day and is worse than no figure at all.
- Accuracy is measured per sensor under load and stated, not quoted from a datasheet.
- Anonymous by construction — overhead and thermal sensing, no faces, no identities, no video leaving the venue.
- The same team integrates occupancy with CCTV, access control, signage and BMS so the number actually changes something.
Limitations & prerequisites
- Occupancy is a balance, so it is only as good as its worst-covered exit. A fire door used as a shortcut will make the figure drift upward until the next reconciliation.
- Counting accuracy degrades as people pass shoulder to shoulder. Published figures of 95–98% assume normal flow; dense surge conditions are lower, and that is when the number matters most.
- Density estimation from an angled camera is less accurate than line counting and should be used as a trend indicator rather than an absolute figure.
- The system measures crowds; it does not manage them. Alerts are only useful where there is a trained team and an agreed response.
- Anonymous counting cannot distinguish a returning visitor from a new one. Unique-visitor figures need a credential-based system such as RFID alongside it.
- Statements here about permitting and civil defence requirements are general guidance, not regulatory advice; confirm the applicable requirements with the relevant authority.
FAQ
Overhead sensors at every entrance and exit of a space count people crossing a line in each direction. Occupancy is the running balance of entries minus exits, reported against that space's declared capacity. Density, flow rate through chokepoints and queue length are measured alongside it, and thresholds raise alerts to operations.
Overhead stereo-vision counters typically achieve 95–98% at normal flow, thermal slightly lower, and overhead LiDAR highest in dense conditions. Accuracy falls as people pass shoulder to shoulder. We measure the real figure per sensor with controlled walk-throughs and state it, rather than quoting a datasheet.
No. Sensors are mounted overhead and see the top of a head and shoulders, or a thermal signature. No facial image is captured, no biometric template is created and no video is retained — the stored record is a time series of numbers.
It depends on the space, the activity and the applicable guidance, and it is set in the event's own safety plan rather than by the system. What the system provides is a measured density figure against the thresholds your safety officer sets, and an alert with enough lead time to act.
It can trigger that. Integration with turnstiles or access control can pause or meter entry at a red threshold and release it as occupancy falls, and signage can redirect visitors to another hall. The decision rules are agreed with the safety officer and every change to them is logged.
For zone density estimation, often yes. For accurate line counting at a doorway, usually not — a face-height camera cannot reliably count a dense crowd, and overhead mounting is what makes counting robust. A typical deployment reuses CCTV for density and adds dedicated overhead sensors at the boundaries.
Counting, occupancy calculation and alerting all continue, because inference and rules run on site. Only remote dashboards and long-term reporting are affected, and data reconciles when the link returns.
Crowd analytics counts anonymously — how many people are in a space right now — and covers everyone including staff and contractors. RFID identifies credential holders and tells you who went where. Venues that need both usually deploy them at the same doorways, where each validates the other.
Discuss your site with an engineer
Tell us the venue, the expected visitor volume and the systems you already run. We reply with a technical view, a realistic scope and the next sensible step — a site survey, a working demonstration, or a full technical and commercial proposal.
Sources & evidence
- UAE Federal Decree-Law No. 45 of 2021 — Personal Data Protection Law — Governs collection, retention and cross-border transfer of visitor personal data in the UAE.
- Dubai SIRA — Security Industry Regulatory Agency — Regulatory framework for security systems and personnel at Dubai venues.
- UAE Government — public event permits — Permitting context for public events in the UAE.
Vendor and product names are trademarks of their respective owners; references are for technical context and do not imply partnership, certification or endorsement.