Calibration and validation of the Legion simulation model using empirical data

A frequently encountered difficulty in the field of pedestrian modelling and crowd dynamics is the scarcity of systematic empirical data. This paper attempts to narrow this gap by discussing the collection and analysis of large amounts of empirical data from around the world, which is used to first calibrate and then validate the microscopic pedestrian interaction model that lies at the heart of the commercial simulation software package ‘Legion’. We briefly review the model which represents pedestrians as learning-adaptive agents with individual preferences and objectives. Next we describe our method of extracting microscopic (individual) and macroscopic (crowd) data from video recordings of real-life pedestrian crowds, filmed at selected locations. The former supply demographic profiles of pedestrian attributes used as inputs, and help calibrate the model parameters by probing context dependencies of the dynamics. The latter yield collective observables and patterns, used to benchmark simulation outputs. We present examples of measurements and corresponding simulations, which demonstrate the breadth and quantitative agreement of the model with the data. Examples include sports spectators queuing as a result of operational procedures; commuters queuing at a bottleneck created by limited vertical circulation capacity; and train passengers boarding, and alighting from, densely populated carriages.