Video: Crowd Simulation Research Aims at Reducing Mass Casualties

This video shows how faculty in UNC's Department of Computer Science are analyzing crowd behavior to predict adverse events like the mass casualties last September in Mecca, Saudi Arabia.

This video shows how faculty in UNC’s Department of Computer Science are analyzing crowd behavior to predict adverse events like the mass casualties last September in Mecca, Saudi Arabia.

By Bradley Allf, Features Writer, UNC’s Institute for Global Health & Infectious Diseases

When someone mentions global health, a lot of people think about disease. Whether it’s medical professionals on the front lines of the latest outbreak of infection or researchers developing a new cancer treatment, when we hear “global health” we tend to think of the medical field. But some people that work to improve global health, such as UNC Professor Dinesh Manocha, PhD, don’t wear scrubs or lab coats. In fact, the only “people” featured in Manocha’s work at all are made up entirely of pixels and binary code. Yet the work Manocha has been doing in the field of crowd simulation is having a big impact on global health.

Manocha, who works in the Department of Computer Science along with his collaborator Professor Ming C. Lin, PhD, and graduate students, aims to better understand how crowds move through space.

“Basically, we are trying to do a lot of mathematical models to see how people behave in different settings,” says Manocha. And it’s a hard problem… but if you have good data, like video data, we can learn from it and then give a better prediction of human behavior.”

By analyzing real-world videos of crowds, he and his team are able to model how a group of people will likely move through a particular space— like a stadium, mall or street-crossing. One need only look back to last September to understand the potential human health benefits of such insight.

It was the last day of the yearly Islamic tradition of Hajj, where millions of pilgrims converge on the Saudi Arabian city of Mecca for a weeklong series of rituals. It is the largest annual gathering of people on earth.

On this particular day, with temperatures soaring near record levels, a mass of people crushed together at the intersection of streets 204 and 223 in Mina, for reasons still unclear. In total, more than 700 people died that day as a result of the “Mina Stampede” as it has come to be called (though some reports suggest the death toll is as high as 2,400). Manocha was contacted shortly afterward by researchers from Saudi Arabia who wanted to know how to prevent such a tragedy from happening again.

This wasn’t the first time Manocha has been approached about using his research to improve crowd safety during Hajj. Since 2011, his group has been working with a team at the Hajj Research Institute of Umm Al-Qura University, Saudi Arabia, to better predict the movement of pilgrims as they perform the various rituals of Hajj. One of their main results was designing a “Virtual Tawaf Simulator,” which can predict the pedestrian flow of pilgrims under different circumstances in the area that surrounds the Grand Kaaba. This was perhaps the very first demonstration of crowd simulation technology being used to predict the movement of tens or hundreds of thousands of people in a dense setting.

By better understanding how crowds move through space, the research Manocha and his team are doing can inform new ideas for improving crowd safety, such as changing the layout of a space or building. That could be a huge step toward making crowds, like those that gather for Hajj, safer.

Dinesh Manocha, PhD,

Dinesh Manocha, PhD, studies video surveillance footage of crowds and then creates models that mimic their patterns of movement.

The Science Behind Crowd Movement
So what does this research look like specifically?

Professor Manocha offers to show me some of the simulations his group is working on. He turns on a computer and pulls up a video showing a crowded city street corner.

“So this is a very famous crossing in Tokyo called ‘Shibuya.’ And this happens at five o’clock. Suddenly, all the lights go red and you see masses of people walking like this, like this– and I’ve been there,” says Manocha. “It’s one of the most congested human crowds.”

He plays the video and I watch an astounding number of people making their way across the street. “People follow their own patterns, they make lanes, and formations. And we’re getting similar patterns here,” Manocha says, pointing to a second video playing beside the first.

This is the simulation. Manocha and his team recreated the physical parameters of Shibuya and then placed one of their simulated crowds into that environment to see how it would move. The pixelated people in the video are obviously computer-generated, but the natural way the crowd moves across the street? Less so. Just like in the real video, each person walks with their own intentionality. Some move quickly, others move more slowly. Some follow the main channels while others carve out a slightly different path. The people even cluster in certain areas in a very similar way in both videos.

Manocha’s team is able to produce such accurate crowd movement patterns because the models that they have designed take into account all sorts of dimensions that affect human movement. These include parameters like collision-avoidance, reaction to the environment, and even personality.

Another such parameter is cultural background. His research has shown that crowd dynamics can change depending on the culture of the people in the crowd.

“Crowds in Asia are very different from crowds in America,” Manocha says. “So we use cultural factors. And other research groups have done experiments, like for example in Asia or in India, there’s not a big notion of personal space. People just move fast and if other pedestrians come close, you may not slow down.” On the other hand, in many Western nations dense crowds move more slowly because those cultures tend to want more personal space.

Such is the breadth of Manocha’s work. He and his team must draw information from all sorts of areas to inform their research because human behavior is governed by more than just a desire to move efficiently from place to place. In another video created by Manocha’s team, one can watch how stress impacts the movement of a crowd. A group of people under a moderate amount of stress is shown to move quickly to its destination. But under a high stress level– the example given in the video is the stress created by a loud alarm– crowd movement breaks down as people clump together, preventing efficient movement.

Such examinations of culture and stress are not the typical bread and butter of computer science research. As Manocha says, “This is a very multi-disciplinary field, from math, to physics, to psychology, to computer vision, to parallel computing, et cetera– all kinds of ideas come in the picture.”

Chapel Hill's Interim Fire Chief Matt Sullivan says he and his team prepare for crowd rush.

Chapel Hill’s Interim Fire Chief Matt Sullivan says he and his team prepare for crowd rush.

Predicting Crowds in Chapel Hill
And this expanded scope of inquiry lends itself to broader applicability. Beyond just Hajj, the research has implications for gatherings of people at any scale. Even the local crowds that gather in Chapel Hill on Franklin Street, say for Halloween or after a big sporting victory, pose significant threats to human health and could thus potentially benefit from some of the insight Manocha’s research holds for improving crowd management.

“I’ve seen crowd rush in events downtown that have been created by firecrackers; I’ve seen it created by guns being shot. I’ve seen it created by really violent and bloody fights.”

That’s Interim Fire Chief of Chapel Hill, Matt Sullivan talking about his experiences coordinating crowd management on Franklin Street. I’m talking with him to try and understand what crowd management looks like on the ground level. According to Sullivan, planning for these crowds is a huge undertaking, requiring months of planning, multiple fire and police departments, and even a weather eye on the performance of the UNC basketball team.

All this is necessary because of the inherent risks associated with such large crowds. Chief Sullivan and his fire team, as well as the police and EMS personnel that make up Chapel Hill’s Crowd Management Team, must contend with all sorts of problems that arise from these crowds: violence, alcohol poisoning, fires and sexual harassment. But one risk they are always acutely aware of is crowd rush. That is, the rapid movement of a mass of people in a crowd.

Unfortunately, in such an event there is little that emergency personnel would be able to do to stop the rush. During crowd rush, emergency personnel are trained not to resist the crowd’s movement.

“When something like that occurs, the responders get out of the way, and there’s specific ways they are trained to get out of the way,” says Sullivan. “They want to get behind a big immovable object– get behind a big tree or in an alcove– and let the rush come by.”

Trying to stop a crowd rush, beyond just being useless, would be dangerous for both responders and the people in the crowd.

Instead, decreasing the dangers of a rush is all about preparation. For example, before the event begins police and fire teams set up barricades around the event perimeter to limit traffic, among other reasons. But they are careful to use only flimsy barricades. Small, lightweight structures are crucial for safe crowd management, as they allow crowds to swell past the barriers should a rush occur.

“We can’t bring in heavy concrete blocks and barriers,” Sullivan says. “We have got to have stuff that can be thrown away pretty easily and quickly so that people can move. So while it has to be secure, if an incident occurred we have got to be able to open up the perimeter to flush people out.”

This pedestrian simulation of a stadium exit was created by Manocha and his colleagues, maintaining the speed and density relationships they observed in real world video of stadium evacuations.

This pedestrian simulation of a stadium evacuation was created by Manocha and his colleagues. It reflects the speed and density relationships they observed in real world video.

This is an idea Manocha himself is interested in. Previously, his team worked on a project where they accurately simulated the movement of a crowd exiting a stadium. One of the next goals of this project is to predict how stadium barricades impact the movement of a dense crowd. Such a project could inform safer policies for utilizing these barricades during crowd events, which might one day change how local police and fire departments deal with the threat of crowd rush.

I am reminded of other ideas Manocha has worked on throughout my conversation with Chief Sullivan. For example, Sullivan says that one important part of being prepared for crowd rush is to have a set of scripted statements that can be sent out over the PA system to control a crowd.

“So a PA won’t say ‘hey you all stop running.’ No, there are scripted messages that they have prepared when something goes bad. It’s like when a tornado is coming you don’t say ‘hey folks a tornado is coming!”’ Sullivan continues, “We’re looking at things you can do operationally not to increase panic.”

As discussed above, Manocha’s research has shown that panicked, stressed out crowds tend to bunch tightly, constricting movement and increasing the dangers of crushing or other crowd disasters. Using calm scripted messages over the PA allows Sullivan and his team to reduce the stress levels in a crowd and thereby decrease the threat it poses.

Thus, the research conducted by Manocha and his collaborators has the potential to impact crowd management policy at multiple scales, from local to global. This is important because the issues associated with crowds will only grow as our world becomes more urban.

Dr. Armin Seyfried is a leading researcher on pedestrian dynamics at the Jülich Supercomputing Centre and a close collaborator with Manocha. He attests to the effects of urbanization on crowd management.

“Worldwide[,] cities are growing. In particular in Europe we are confronted with the problem that the process happens in the context of a historical urban structure with often limited space,” Seyfried says. “More and more we have problems that there is no space to expand transport infrastructures…”

Seyfried provides historical city centers, in which modification or removal is typically out of the question, as an example of one of the factors limiting the infrastructure changes cities can put in place to accommodate growing crowds. As our urban centers expand and begin pushing against the walls of this limited space, a better understanding of crowd dynamics will be vital to ensuring that these areas are safe.

For Manocha, the application of his research to real-world problems that pose risks to human health is the most rewarding aspect of the work.

“If our research results can be used to improve the lives of others, that is one of the major fulfilling aspects of this research,” he says. “My entire team of students and collaborators would love to develop technologies that can be used by others, or used to improve the lives of others by designing better spaces for pedestrian flows and reducing the chances of a crowd disaster.”

So while he may not study medicine or public health, Manocha’s work is undoubtedly tackling a serious global health issue, and it has the potential to lead to safer communities—from Chapel Hill to Mecca.

Leave a Reply

Your email address will not be published. Required fields are marked *