Verification and Validation in pedestrian dynamics
Verification and validation are important concepts in the field of pedestrian dynamics, which studies the movement and behavior of pedestrians in crowded environments. Verification refers to the process of ensuring that a model or simulation accurately represents the real-world system it is intended to model. Validation, on the other hand, is the process of evaluating the model or simulation against real-world data to determine its accuracy and usefulness. Together, verification and validation are used to assess the quality and reliability of pedestrian dynamics models and simulations.
In the pedestrian dynamics community, standards have been established for programs that simulate the movement of pedestrians. These standards were created by conducting small-scale experiments in which the movement of pedestrians was recorded. A typical example of such an experiment is a corridor of a specific width and length where two groups of pedestrians (of a certain size) have to switch positions. It must be demonstrated that the flow of the simulated pedestrians is similar to the flow of real pedestrians, and that this process must occur within a certain time frame. Software and simulators must be able to reproduce the results of these established scenarios. The most widely used standards in this community are the NIST technical note 1822 , its expansions, and the IMO Guidelines . Research has demonstrated that our Crowd Simulation engine complies with these standards. Evacuation In evacuation studies, researchers compare the time it takes for people to evacuate a building in real life to the time it takes in a computer simulation. Researchers use the results of these comparisons to create software that can predict how people will move during an evacuation. For example, in one study, researchers used a computer program to simulate the evacuation of small groups of people and found that it matched the real-life evacuation in terms of how long it took and how people moved. The research showed, when small groups of people evacuate together, it takes longer than when individuals evacuate alone. In our research paper , we showed that our simulations reproduce these findings.
Multiple models have been developed to prevent pedestrians from colliding with each other and with obstacles in the environment. One of the first models developed is the Social Force Model (SFM) , which is still used by several crowd simulation programs. However, SFM has limitations in dealing with complicated scenarios, such as handling multiple criteria and high crowd densities, which may result in unrealistic behavior. Another category of collision-avoidance methods is the (proactive) vision-based method . This category yields more realistic collision avoidance behaviors than the SFM because these are more based on how people see the environment and react to anticipated collisions. Such methods aim to minimize the spent energy, as is often observed in human motions. The model yields desired emergent behaviors, such as lane forming and wave forming. Even more powerful are methods based on Reciprocal Velocity Obstacles (RVO) , which are being used as a default by SimCrowds. This approach solves high-density situations well in which different groups of pedestrians need to swap positions in a narrow passage or navigate around corners of obstacles. Other methods usually fail to solve such problems. SimCrowds allows you to use this method in designated areas where different collision-avoidance algorithms do not lead to realistic flows. In our latest version, we improved RVO so that this method can better handle non-moving (static) pedestrians, as is demonstrated in the video below by the crowd filling up a concert space.
The validity of the simulations can be checked by comparing the trajectories of real humans with the simulated trajectories. For instance, so-called fundamental diagrams can be used to (often visually) compare the simulated data with the real data .
Measuring Data and Metrics
In the research community, Fruin’s Level of Service (LODs)  is used to measure the relationship between crowd density and danger level, which is visualized on a map. However, using only one map is insufficient because different norms apply in different settings. For example, the danger norm at a train station may be two pedestrians per square meter during a specific time, while this threshold would not be considered dangerous in an area where a larger audience is viewing a concert. While using different tables can lead to better conclusions, there is not yet a consensus in the community about the perfect way to measure crowd densities. Traditionally, these measurements are done using a grid, Voronoi diagram, or Gaussian kernel , but these methods can be affected by choices such as whether to include or exclude obstacles and which distance metric to use.
The latest developments in this field focus on validating simulation models in real-time. We were part of a small consortium that created a setup of 32 3D cameras at a train station in London to map the flow of pedestrians in real-time. These cameras measure the number of pedestrians crossing a specific area and their directions. This information is used to automatically calculate an origin-destination matrix, which is continuously updated. Such a system can be used to make predictions of future crowd states. Typically, they have three minutes of computation time to predict what will happen in the next 15 minutes. Based on the results, we can make changes in real-time to optimize the flow of pedestrians and create a safer and more comfortable experience for travelers.
Do it yourself
If you’re looking to take your crowd simulation game to the next level, give our software SimCrowds a try! Our team is always here to help and answer any questions you might have. Don’t hesitate to reach out to us via our website or email. Let’s work together to make your simulation goals a reality!
 E. Ronchi, E. D. Kuligowski, P. A. Reneke, R. D. Peacock, and D. Nilsson. “The process of verification and validation of building fire evacuation models. US Department of Commerce”. National Institute of Standards and Technology, 2013.
 International Maritime Organization. “Guidelines for evacuation analyses for new and existing passenger ships, MSc/circ.1238”, 2007.
 A. Kremyzas, N.S. Jaklin and R. Geraerts. “Towards Social Behavior in Virtual-Agent Navigation”. Science China - Information Sciences, 59:112102, pp. 1-17, 2016.
 D. Helbing and P. Molnár. “Social force model for pedestrian dynamics”. Physical review E, vol 51, no. 5, pp. 4282-4286, 1995.
 M. Moussaïd, D. Helbing and G. Theraulaz. “How simple rules determine pedestrian behaviour and crowd disasters”. Applied physical sciences, vol 108, no. 17, pp. 6884-6888, 2011.
 J. van den Berg, S.J. Guy, M.C. Lin and D. Manocha. “Reciprocal n-body collision avoidance”. Robotics Research, pp. 3-19, 2011.
 U. Chattaraj, A. Seyfried and P. Chakroborty. “Comparison of pedestrian fundamental diagram across cultures”. Advances in Complex Systems, vol. 12, no. 3, pp. 393–405, 2009.
 J. J. Fruin. “Pedestrian planning and design”. Technical report, 1971.
 A. Hillebrand, H. Hoogeveen and R. Geraerts. “Comparing different metrics quantifying pedestrian safety”. In Pedestrian and Evacuation Dynamics, 2018.
#Pedestrian #Dynamics #Verification #Validation #Modelling #Crowd #Simulation
Read, like and share this post on our LinkedIn!