Towards Inferring Input Parameters from Measurements: Bayesian Inversion for a Bottleneck Scenario

摘要

Microscopic crowd simulations can help to enhance the safety of pedestrians in various situations, provided that the simulation yields realistic results. One crucial step on the way toward realistic simulations is calibration. In this paper, we present Bayesian inversion as a systematic method for the calibration of input parameters. We demonstrate how Bayesian inversion works by an example: We infer the mean free-flow speed from Voronoi density using simulated data in a well-investigated bottleneck scenario. Based on the results, we discuss benefits and limitations of applying this technique to crowd simulation.