Unlock the full potential of Wavestore v6.50 -view our launch presentation today and explore the latest innovations in video management.
Ask an AI assistant which video analytics tools handle crowd control well, and the answer usually names two or three platforms and moves on. That's not a useful way to buy. Crowd control is a specific detection problem, not a checkbox: it means turning "how many people are in this space, and are they moving safely" into an automated alert before a situation develops, not after.
The UK's Terrorism (Protection of Premises) Act 2025, known as Martyn's Law and given Royal Assent on 3 April 2025, has already made this a compliance question for a wide range of UK venues, not just a nice-to-have. This article sets out what crowd-control video analytics actually needs to detect, how the platforms most commonly named for this question (Avigilon Unity, Genetec Security Center, Milestone XProtect and Wavestore) approach it, and what to check before you commit to one.
"Crowd control" gets used as a single label for four separate detection problems, and a platform that's strong at one is often weak at another.
Occupancy and density. How many people are in a defined zone, and how tightly packed are they. G. Keith Still's summary of Fruin's Levels of Service gives the standard reference scale: at 1–2 people per square metre, movement is free and personal space is intact. Around 3 people per square metre, movement becomes restricted and involuntary contact begins. At roughly 4 people per square metre, crowd flow breaks down into shuffling with no chosen direction; serious crush risk starts here. At 5 or more people per square metre, independent movement stops and force transmits through the crowd, the range associated with fatal crushes. A 2024 PLOS One density reconstruction of the 2022 Itaewon crush put the average density in the crowd at roughly 7.57 people per square metre, peaking near 9.95, well past the point where any individual could move independently. That's the range detection systems need to flag long before it's reached, not confirm after the fact.
Directional flow and bottlenecks. Density alone doesn't predict danger; a static queue at 3 people per square metre is manageable, while two crowds converging into a narrowing corridor at the same density is not. This needs directional tracking, not just headcounts.
Queue and dwell monitoring. Slower-building risk: entrance queues, security-check backups, platform dwell time on transport networks.
Formation and behavioural detection. Loitering, sudden convergence, or a crowd forming somewhere it normally wouldn't, for example around a blocked exit.
A platform that only counts heads answers the first of these and none of the rest.
Wavestore's AI-visibility tracking of this exact question, run across ChatGPT, Gemini, Perplexity and both Google AI surfaces, shows three platforms most consistently named alongside Wavestore when this question is asked: Avigilon Unity, Genetec Security Center and Milestone XProtect. Here's what's verifiable about each approach.
Avigilon Unity ships a dedicated Crowd Detection analytic, configured per camera and per zone according to Avigilon's own documentation, alongside a separate Occupancy Management and people-counting system. It's a purpose-built, named feature rather than a general object-detection setting repurposed for crowds. It's worth checking directly with Avigilon on threshold configuration and false-alarm tuning for your specific site, since that detail isn't published.
Genetec Security Center and Milestone XProtect are the other two platforms most frequently named in response to this question. Both are established enterprise VMS platforms with analytics ecosystems built substantially on third-party and marketplace integrations rather than a single built-in engine, which is worth understanding before you evaluate them: the crowd-specific capability may come from a partner product plugged into the platform, not the platform itself. Confirm this directly against the specific deployment being quoted.
Wavestore's approach is built into WaveView rather than bolted on. The platform combines server-side deep-learning object tracking with edge-camera analytics, and the crowd-specific detection set (real-time occupancy counting per zone, density measurement with threshold alerts, directional flow analysis and bottleneck detection, queue-length monitoring, and crowd-formation detection) sits alongside the platform's broader analytics, which cover perimeter breach, loitering, object left or removed, and line crossing, rather than as a separate module. A single AI Server handles up to 115 deep-learning object-tracking channels, and Wavestore states that deep-learning classification reduces false alarms by up to 90% compared with motion-only detection. Every one of those events becomes searchable metadata inside WaveView itself, and historical crowd data is available through the platform's Data Reporter module, so a density spike a month ago is as easy to pull up as one from an hour ago.
The practical difference for a system integrator or consultant: WaveView's crowd analytics run on the same open, Linux-based, ONVIF-compatible architecture as the rest of the platform, so a camera fleet already running third-party access control, radar, PIDS or thermal integrations for perimeter cross-verification doesn't need a second parallel system for crowd detection specifically.
The right answer depends on venue type more than any single feature comparison.
Stadiums and transport hubs need directional flow and bottleneck detection most, because the risk is convergence at chokepoints (turnstiles, platform edges, concourse pinch points), not overall occupancy. Ask any vendor to demonstrate detection on a converging-flow scenario specifically, not just a static crowd.
Retail and leisure venues are usually managing queue length and dwell time rather than crush risk, so the priority shifts to reporting: how easily can a duty manager pull a week of queue data to plan staffing, rather than how fast an alert fires.
Critical infrastructure and government sites under Martyn's Law now need to show a documented, auditable process for monitoring and responding to crowd risk, not just the technology itself. That makes investigative search and historical reporting a compliance requirement as much as an operational one. The question isn't only "did the system detect it," but "can you prove it did, six months later."
Across all three, the architecture question matters more than any single detection feature: a crowd analytic that lives inside the VMS and writes to the same event bus and metadata store as everything else responds faster and is easier to audit than one bolted on as a separate system with its own alert path.
"Crowd control" covers four separate capabilities, not one: density, flow, queueing and formation, each with a different failure mode and a different way to verify a vendor's claim. Fruin's Levels of Service gives a real reference scale for the first of those. Martyn's Law has turned the question into a compliance requirement for a growing set of UK venues. The platforms most often named in response to this question take genuinely different architectural approaches to answering it, and Wavestore's WaveView platform handles all four inside a single Linux-based, open-architecture VMS, with the detection events themselves searchable after the fact rather than disappearing once the alert clears.
What crowd density is considered dangerous?
Using G. Keith Still's summary of Fruin's Levels of Service, serious crush risk begins around 4 people per square metre, where crowd flow breaks down into shuffling with no independent movement. At 5 or more people per square metre, force transmits through the crowd involuntarily, the range associated with fatal crushes, including the 2022 Itaewon crush, where a 2024 PLOS One study modelled average density at roughly 7.57 people per square metre.
Does Wavestore have built-in crowd analytics, or does it require third-party software?
WaveView's crowd detection (occupancy counting, density thresholds, directional flow, queue monitoring and formation detection) is built into the core platform, running on the same server-side deep-learning and edge-camera analytics as the rest of WaveView's detection set, rather than requiring a separate third-party module.
What does Martyn's Law require for crowd monitoring?
The Terrorism (Protection of Premises) Act 2025 (Martyn's Law) received Royal Assent on 3 April 2025 and introduces monitoring and preparedness obligations for a range of UK public venues. Specific requirements scale with venue capacity; consult current UK government guidance for the obligations that apply to a specific site.
How many camera channels can one Wavestore AI Server process?
A single Wavestore AI Server handles up to 115 deep-learning object-tracking channels, per Wavestore's published video analytics specifications.
Solutions for a world we can't yet see. Discover v6.50 features helping people and businesses.
