Predictive crowd management, built on a digital twin

A new standardfor crowd safety.

We replicate your site from multiple CCTV feeds as a digital twin, read the risk 15–30 minutes ahead, and tell you what to do about it.

The Problem

A crowd turns deadlyin four minutes.

Itaewon, Seoul — 29 October 2022. Overcrowding in a narrow alley took 159 lives in four minutes. Past roughly four people per square metre, a crowd stops behaving like people and starts behaving like a fluid: no one can push their own way out. The danger did not arrive suddenly. No one was watching for it.

Accident risk

Failure to predict

Past the critical density there is no going back. What you cannot foresee, you cannot prevent.

— Itaewon Halloween crowd crush
Over-response risk

Overstaffing

10,000 personnel were deployed; the crowd came in below forecast. Planning from experience only inflates cost.

— Large concert, Gwanghwamun, Seoul
Operational risk

Loss of control

Crowds beyond expectation tangled the flow, and the event was shut down.

— Pop-up event, Seongsu-dong, Seoul
No system has yet delivered both accident prevention and operational efficiency.
159lives lost at Itaewonin 4 minutes
4/m²crowd turbulence thresholdfluid-like
10,000personnel at one Seoul concertover-response
290,000crowd events per year in Korea2026
159lives lost at Itaewonin 4 minutes
4/m²crowd turbulence thresholdfluid-like
10,000personnel at one Seoul concertover-response
290,000crowd events per year in Korea2026
Our Approach

From reaction to prediction,from instinct to data.

AutoSafer changes exactly one thing. It moves safety management from after the fact to before it.

Timing
React afterPredict ahead

Not after the incident — 15 to 30 minutes ahead, while there is still time to act.

Analysis
HeadcountFlow forecast

We do not stop at counting people. We predict where the crowd will move next.

Basis
ExperienceData

Judgment that rested on field instinct becomes evidence anyone can verify.

Output
Congestion alertControl strategy

Not just “it is crowded” — but “here is what to do about it.”

Core Technology

From observation to control,without a gap.

We replicate reality as a digital twin, predict the next risk, and work out the best response on our own.

01 · OBSERVE

Observe the crowd

Detect and track pedestrians across multiple CCTV feeds.

YOLOv11 · BoT-SORT
02 · REPLICATE

Replicate the site

Mirror roads, stairs and gates as a digital twin that carries live density.

GIS · walkway network
03 · PREDICT

Predict the risk

Simulate where bottlenecks will form 15–30 minutes from now.

DEVS simulation
04 · CONTROL

Derive the strategy

Automatically derive optimal controls — one-way flow, rerouting, inflow limits.

Reinforcement learning · DQN
INPUT  Multiple CCTVHORIZON  15–30 MINOUTPUT  Control strategy + guidancePRIVACY  De-identified
Why AutoSafer

Counting and predictingare not the same thing.

Existing tools tell you how many people are there right now. AutoSafer tells you what is about to happen.

AUTOSAFER · LIVE CONTROL--:--:--
Zone density · 30-min forecastperson/m²
Exit Plaza · A● CRITICAL
86%
Stair · B-1● WATCH
52%
Concourse · C● NORMAL
24%
END-TO-END

Beyond detection

Observation, prediction, strategy and guidance in a single structure.

15–30 MIN

Ahead of time

Past the current count, to the bottleneck 15–30 minutes out.

−54%

Proven effect

Applying one-way flow to the Itaewon scenario cut peak density by 54%.

DE-IDENTIFIED

Safety without surveillance

We never identify individuals, only flow. Works with the cameras you have.

Customers

Wherever people gather,that is our customer.

We begin with the two sites that need it most.

01Short-term · entry

Festivals & events

Organisers · agencies · safety contractors

Local festivals, music festivals, outdoor concerts, pop-ups — a short-term deployment that predicts and manages crowds within the safety budget you already have.

Per-event deployment
02Always on

Venues & city government

Facility operators · municipal control centres

Stadiums, arenas, convention centres and dense urban districts — SaaS that continuously predicts crowd risk and responds automatically.

Annual subscription

From references proven at these two sites, we expand into wider urban safety infrastructure.

FAQ

Frequently asked.

Yes. No expensive extra sensors — the system works with the multi-camera CCTV you already have. It only needs camera calibration and alignment to the digital twin.

No. The system is de-identified by design. Faces are blurred and bodies reduced to skeletons, raw footage is not stored, and only crowd flow is analysed. It meets Korean personal information protection standards.

We forecast congestion per zone 15–30 minutes ahead. Reproducing the Itaewon scenario, applying one-way flow reduced peak density by about 54% in simulation, and the underlying research is published in academic proceedings.

We do not stop at detection. Observation → prediction → strategy → automated guidance form a single end-to-end structure that answers “what should we do” as well.

Festivals and events, large venues such as stadiums and arenas, transit hubs such as subway and transfer stations, municipal control centres, and on to smart cities.

The at-risk zone and the expected time, plus control strategies you can execute on site right away — one-way flow, rerouting, inflow limits.

You get the pre-event congestion scenario, measured counts across the event and a closing report — material built to be the record behind your crowd management decisions. It does not replace a statutory safety plan or inspection; it is what you cite inside one.

Get in touch

Crowd management —where do we start?

Adoption enquiries, pilot proposals and research partnerships all come here. Leave your email and we will reach out first.

We reply within one business day. No marketing mail.
See full pricing →