Validation of Pedestrian Groups in Agent-based Simulation

We develop a unique validation method to empirically assess the influence a group model. We apply a computer vision pedestrian trajectory analysis and adapt it to crowd simulation. We determine the probability that collocated virtual pedestrians are travelling in the same group based on their path over time. We use receiver operating characteristic (ROC) analysis to determine the best parameter intervals. A series of experiments is run in a crowded corridor of a virtual airport. The results demonstrate that our method is able to provide crowd modelers a unique description of the influence of the model parameters.