An Artificial Neural Network Framework for Pedestrian Walking Behavior Modeling and Simulation

Abstract

Movement behavior models of pedestrian agents form the basis of computational crowd simulations. In contemporary research, a large number of models exist. However, there is still no walking behavior model that can address the various influence factors of movement behavior holistically. Thus, we endorse the use of artificial neural networks to develop walking behavior models because machine learning methods can integrate behavioral factors efficiently, automatically, and data-driven. In this paper, we support this approach by providing a framework that describes how to include artificial neural networks into a pedestrian research context. The framework comprises 5 phases: data, replay, training, simulation, and validation. Furthermore, we describe and discuss a prototype of the framework.