Every year people die when panic breaks out in crowds — in settings ranging from stadiums to amusement parks to train stations. "The victims suffocate or are trampled to death," says Dr. Gerta Köster of Siemens Corporate Technology (CT) in Munich, Germany, describing the most extreme cases. Köster and her ten-person team are working on making busy places safer by simulating crowd flows, using a method known as "crowd control." Team members simulate to what degree certain scenarios, involving different crowd densities, affect the movement of a group as a whole. "Thanks to tremendous advances in computing power, there has been intensified research in this area worldwide for about ten years now," reports Köster.
Simulation and Virtual Reality
Crowd Flow Simulation: Predictive Vision
Crowd flow simulations will enable us to make public places safer. The technology makes it possible to recognize potentially dangerous crowd congestion and take appropriate measures — before an incident occurs.
Optimizing Flow Dynamics at Events.
CT’s Crowd Control team has expanded on work that began as basic research at the university level. The idea is to apply the resulting knowledge to improve solutions at several Siemens Groups — for example at the Mobility Division — in order to improve the design of products such as train interiors and thus reduce the time needed for passengers to board and disembark trains in stations. Plans also call for crowd flow simulations to be applied to logistical planning of events, infrastructures, and transportation systems at Siemens’ Building Technologies Division.
In her lab, Köster’s computer displays a range of colors — green represents low crowd density, yellow is medium density, and red stands for high density. If an area is spacious, people tend to keep a good distance from one another and from walls. But as soon as people approach a narrow passageway, they change their behavior, moving closer to one another and pressing closer to the walls as they try to advance through the bottleneck. This is how crowd congestion develops. When they once again have more space, people move away from one another — and the dark red color reverts to yellow and green.
Simulating the Behavior of Thousands of Pedestrians.
"We can simulate and visualize the behavior of tens of thousands of pedestrians moving simultaneously," says Köster. "Our simulations are based on a special mathematical approach known as cellular automata." The technique involves using a hexagonal grid to divide a 2D space into cells, each of which must have an unambiguous state: "empty" or "occupied." If a cell is occupied, this means it is occupied by a person, an obstacle, or a destination. The status of the cell is continuously and automatically updated according to set rules. A program calculates how pedestrians approach a destination, how they interact with one another, and how they maneuver around an obstacle. "This is similar to the movement of electrons," says Köster. "Destinations attract, obstacles repel, and people keep a certain distance from one another."
To better depict human behavior, Köster’s research team has gradually refined its computer model. Instead of considering the age, gender, and fitness of each person individually, the effect of the program’s parameters is collectively allocated to pedestrian movements. This makes it possible to reproduce a fundamental relationship with great accuracy. The denser a crowd becomes, the slower it moves. "We have succeeded in keeping the algorithms lean by combining, in a very concentrated way, many factors that affect the behavior and movement of pedestrians. This is why our simulator can run such fast calculations," explains Köster.
Decision-Making Aids for Event Managers.
Speed is particularly important when using the simulator to make short-term forecasts. The goal is to provide event managers with decision-making aids before an incident occurs. In a scenario in which a large crowd is waiting on a train platform after a concert or soccer game, for instance, and a delayed train packed with passengers is due to arrive at the station, the control center supervisor can call for a short-term forecast by just pushing a button. With a warning from Köster’s simulation three minutes before life-threatening crowd congestion forms, an extra train might be rushed into service.
"Our simulator calculates such a scenario in time lapse motion. This means that it is ten times faster than real time when dealing with 5,000 pedestrians," says Dr. Wolfram Klein, who played a major role in programming the simulator. The speed of the simulation tool is unique worldwide. It is clearly superior to that of simulators based on a process known as the "multi-agent approach," which requires hours to run. Although the multi-agent algorithm can determine positions of individual people more precisely, Köster points out that "we are more concerned with being able to calculate crowd density as quickly as possible." Siemens has already submitted patent applications for eight associated inventions.
The mathematical approach employed by Siemens was first used by the theoretical physicists Prof. Kai Nagel and Prof. Michael Schreckenberg in their "Nagel-Schreckenberg" model for traffic simulation. Siemens researchers have refined the model in collaboration with Prof. Ernst Rank’s Department of Computation in Engineering at the Technical University of Munich.
Köster’s team plans to develop a simulator to process real time data, collected with camera systems and wireless technologies. The researchers want the simulator to accept data online and then filter out essential information so that it can be added to an algorithm — a prerequisite for fast simulation calculation.
Simulating Topographical Data and Groups within Crowds.
This area of research is an important part of a project called "Regional Evacuation: Planning, Control and Adaption" (REPKA), which is sponsored by Germany’s Federal Ministry of Education and Research. Siemens, the Technical University of Munich, the Kaiserslautern police, and the Fraunhofer Institute for Integrated Circuits in Erlangen are participating in the project. REPKA monitors and analyzes crowd flows after soccer games played in the Fritz-Walter stadium in Kaiserslautern. The project’s goal is to provide better, more flexible planning and control of stadium evacuation into an area with a radius of 1 km.
Köster’s job is to develop the evacuation model such that the movement of 50,000 pedestrians can be calculated in real time — without the need for powerful computers. At the same time, information on the behavior of groups of people, such as fans or families, who also interact with one another, is to be incorporated. Topographical conditions around the stadium are also part of the model, since the facility is on an elevated site right in the middle of a densely built-up residential area. All this must work not only with high precision but also in real time. "These are major challenges," says Köster. "We have to improve our models, test them, and adapt the software code so that it will still be very fast."
REPKA includes plans for a virtual training simulator, which by 2011 will enable operations leaders to run through rescue operations and see the impact of their decisions. "We want to give people experience with situations that can’t be tested in real life," explains Köster.
The scenarios are to show what would happen if stadium exits became blocked because of fire, how crowds would respond to public announcements, in which directions they would run, and where dangerous crowd congestion could develop. Plans also call for the training simulator to show in real time what happens if an emergency exit is opened. "We add these interactive factors to our crowd flow simulation. And that puts us once again on new terrain in this field of research," says Köster.