Microscopic calibration and validation of pedestrian models — Cross-comparison of models using experimental data

Abstract

Calibration of microscopic simulation models is complex due to amongst other things the large number of model parameters, and the lack of suitable data. Rather than a qualitative test to determine whether model predictions are plausible on an aggregate (macroscopic) level, this contribution focuses on statistical calibration of model parameter from individual pedestrian trajectory data. To this end, a new generic approach is presented enables parameter estimation for microscopic models in general and in particular for walker models. The application results provide new insight into the behavior of individual pedestrians, the inter-pedestrian differences, as well as the resulting pedestrian flow characteristics. By comparing different models of increasing complexity, it is investigated which of the model amendments are significant from a statistical point of view and which are not. It is shown that besides anisotropy, finite reaction times play an important role in correctly describing microscopic walking behavior.