Could Crowd Simulation Software Have Prevented the Seoul Disaster?

Or, technology aside, could it be as simple as teaching people manners (Don’t push)?

The recent disaster in Seoul, South Korea, where 150 people were crushed to death in alleys, has raised many questions. Chief among them: How could this have happened?

A nighttime crowd estimated at around 100,000 had gathered in Seoul’s Itaewon district. Once popular with U.S. military personnel stationed there during the Korean War, Itaewon has since become popular with Koreans and tourists alike due to its many restaurants, bars and stores—and its vibrant nightlife. Like everywhere else, Itaewon suffered during the Covid-19 pandemic. But that Saturday (October 29), with the mask mandate recently rescinded, it was party on. And what better place to celebrate than Itaewon? It was doomed to end badly, though, when people on both ends of an alley tried to push their way through.

Instagram star Sun Seo-jung tells of a scene with people yelling, “Hey, push!” “We’re stronger!” and “I’ll win!” before blacking out from being squeezed from the front and back. The result: over 150 deaths.

Crowd surges have been responsible for many mass death events. There have been concertgoers crushed against the stage (Guns N’ Roses concert in the UK, 1988), soccer fans pressed against chain-link fences (Orkney Stadium disaster, South Africa, 1991), and pilgrims at annual religious gatherings in India and Saudi Arabia (in the central Indian state of Madhya Pradesh in 2013 and at a hajj pilgrimage in 2015, respectively), to name just a few. The 2015 Mina disaster killed over 2,400 pilgrims in Saudi Arabia, making it the deadliest hajj in history and possibly the worst crowd disaster of modern time.

A study of 100 mass death incidents resulting from crowds between 2006 and 2016 adds up to 4,011 deaths and more than 6,346 people injured.[i]

That such a crowd disaster occurred in South Korea, a nation usually adept at crowd control, is somewhat surprising. When K-pop sensation BTS held a concert attended by 55,000, it was policed by 1,300 officers. Yet only 137 police officers were on duty in Itaewon that Saturday night with orders to only be on the lookout for sexual harassment, theft and drug use—not for crowd control, according to the New York Times.

No Way Out

The alleys in Itaewon where people were killed were open to streets—not dead ends or fenced-in areas. The study of fluid behavior (gas or liquid) may have one thinking that people would behave the same way and that people would have spilled out onto the bigger street, relieving the pressure, as it were. But people tend to behave more like particles than a continuous media, and as such, are quite prone to get stuck as their numbers go up.

In a 2018 study,[ii] researchers modeled the problem as if people were similar to particles of gas in a cylinder, with a moveable piston that causes the particles to be closer together, and thereby exert forces on each other. Applied to the Itaewon disaster, the people doing the pushing take the place of the piston.

Complicating the matter even further is crossflow. From the account of the survivor above, crowds from two streets were pushing to get to the other side, trapping people in the middle with no way out.

Cause of Death

Crowds have been known to trample over others that fall in front of them, and while that can obviously be most unpleasant, painful and result in injury, rarely does the weight of people walking or running over a body cause death. The far more common cause of death in a crowd is suffocation, or compressive asphyxia.

People die standing up, not lying down—or, in extreme cases, they are lifted off the ground but still remain vertical. They are pressed together or against a wall or fence with the collective force of many people who are pushing toward them.

A slow asphyxiation will result if your lungs contract from exhalation and because of external forces, you cannot inhale. It is the way a boa constrictor would kill an adult, constricting their chest and preventing expansion. The snake will not crush the air out of you. It’s not strong enough do that, but it can take up the slack when you exhale and prevent you from inhaling.

Why Isn’t There an App for This?

Crowd simulation such as this open-source application seems to model orderly crowds. (Picture courtesy of Vedere.)

Crowd simulation such as this open-source application seems to model orderly crowds. (Picture courtesy of Vedere.)

With bad crowd behavior as old as crowds themselves and in the age of technology that can seemingly save us from every disaster, can’t we reasonably expect software to provide a solution to this problem as well?

There are many software products that help model crowds and pedestrian traffic. Bentley Systems acquired LEGION software years ago for this purpose. LEGION is sophisticated and 3D, and it is able to show simplified versions of people (dots) as well as 3D (humanoids) and can simulate the dynamics of people moving through and around buildings and structures. But modeling the worst crowd behavior, that which causes death, remains elusive.

In 2016, Mitsubishi Electric and the University of Tokyo’s Research Center for Advanced Science and Technology (RCAST) reportedly developed “the world’s first real-time crowd-congestion estimation system.” Instead of relying on historical data, photos or body counts from events and foot traffic, the software analyzed images from CCTV cameras and claimed that it was 80 percent accurate in estimating crowd congestion, superior to crowd control simulators, which relied on historical data and had only a 50 percent chance of predicting congestion. (It is not known if Mitsubishi ever developed the camera-based system into a commercial product.)

A Little Education Wouldn’t Kill You

The push of using a cylinder to simulate the push of people into each other seems to offer an obvious cause to the forces that result in mass deaths from overcrowding. But unlike a simple mechanical explanation, the lethal activity of human crowding defies a neat engineering solution. Simulation software may be able to predict the forces that can lead death by asphyxiation, but how could it predict when that would occur, especially with spontaneous events of unorganized celebrations like the one that preceded the Itaewon disaster?

Despite a sickeningly long list of mass crowd death events that starts from ancient times, crowd events resulting in mass death still occur relatively infrequently after long periods of events that occur without incident. While more foot traffic control is an obvious answer and has been shown to work with scheduled events, a technical answer to crowd control for spontaneous events—or scheduled events with an unforeseen incident (a bomb, fire or other panic-inducing event)—is still elusive.

It may be a knee-jerk public reaction to round up the drunk and disorderly pushers caught on camera at the Itaewon disaster. While they may have been behaving badly, they had no murderous intent. They had no idea that, collectively, they were applying lethal force. They couldn’t see or hear the victims.

Perhaps more effective in preventing mass death by crowds would be public education, such as public service campaigns, signage and messages that suggest how one should act in a crowd. The several accounts of mass death attributed to “reasons unknown” indicate that the general public and even police may not realize that the simple act of pushing into a crowd can have lethal consequences.

Preventing most mass deaths caused by crowd surges could be as simple as teaching people one manner: don’t push. Lessons can be supported by simple theory, a basic engineering principle. A little force applied over a large area can result in a large force. It is what makes cars go up in lifts when your oil gets changed. Simple, right?


[i] Cai M.  “The Research and Application of The Crowd Density Analysis and Risk Early Warning of Stampedes Occurred in Open Public Places.” Master’s thesis, Jinan University, 2017.

[ii] “Exploring the Consequences of Crowd Compression Through Physics-Based Simulation.”Libo Sun1, and Norman I. Badler2, Sensors, NIH Library of Medicine, December 2018, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6308419/#B1-sensors-18-04149.