Building crowd movement model using sample-based mobility survey

Crowd simulation is a well-studied topic, yet it usually focuses on visualization. In this paper, we study a special class of crowd simulation, where individual agents have diverse backgrounds, ad hoc objectives, and non-repeating visits. Such crowd simulation is particularly useful when modeling human agents movement in leisure settings such as visiting museums or theme parks. In these settings, we are interested in accurately estimating aggregate crowd-related movement statistics. As comprehensive monitoring is usually not feasible for a large crowd, we propose to conduct mobility surveys on only a small group of sampled individuals. We demonstrate via simulation that we can effectively predict agents' aggregate behaviors, even when the agent types are uncertain, and the sampling rate is as low as 1%. Our findings concur with prior studies in urban transportation, and show that sampled-based mobility survey would be a promising approach for improving the accuracy of crowd simulations.