Modeling crowds as single-minded entities

Inspired by Gustave Lebon's idea of crowds as single-minded entities, we present a novel approach to describe the behavior of a crowd as a single entity, based on the global movement of the entire aggregate of people conforming the crowd. The present work significantly differs from existing literature where the behavior of single individuals within the crowd are the building blocks to describe crowd behavior. A bidimensional neural gas network is implemented to learn the topology of the physical environment in an unsupervised fashion, then a self-organizing map and a Bayesian network are used to describe the behavior of the crowd as a single entity. Experiments were conducted using footage from New York Grand Central Station to test the accuracy of our model to learn and identify different behaviors of the crowd. Results show high accuracy to identify behaviors under usual circumstances and low but consistently increasing accuracy over time on less common cases.