Modeling Live Crowd Emotion Dynamics for State Estimation and Prediction

Crowd violence and the repression of free speech have become increasingly relevant concerns in recent years. This paper considers a new application of crowd control, namely, keeping the public safe during large scale demonstrations by anticipating the evolution of crowd emotion dynamics through state estimation. This paper takes a first step towards solving this problem by formulating a crowd state prediction problem in consideration of recent work involving crowd psychology and opinion modeling. We propose a nonlinear crowd behavior model incorporating parameters of agent personality, opinion, and relative position to simulate crowd emotion dynamics. This model is then linearized and used to build a state observer whose effectiveness is then tested on system outputs from both nonlinear and linearized models. We show that knowing the value of the equilibrium point for the full nonlinear system is a necessary condition for convergence of this class of estimators, but otherwise not much information about the crowd is needed to obtain good estimates. In particular, zero-error convergence is possible even when the estimator erroneously uses nominal or average personality parameters in its model for each member of the crowd.