A Model of Human Crowd Behavior :
Group Inter-Relationship and Collision Detection Analysis

R. Musse and D. Thalmann {soraia,thalmann}@lig.di.epfl.ch Computer Graphics Lab. Swiss Federal Institute of Technology EPFL, DI-LIG, CH 1015 Lausanne, Switzerland

Abstract

This paper presents a model of crowd behavior to simulate the motion of a generic population in a specific

environment. The individual parameters are created by a distributed random behavioral model which is

determined by few parameters. This paper explores an approach based on the relationship between the

autonomous virtual humans of a crowd and the emergent behavior originated from it. We have used some
concepts from sociology to represent some specific behaviors and represent the visual output. We applied our

model in two applications: a graphic called sociogram that visualizes our population during the simulation, and

a simple visit to a museum. In addition, we discuss some aspects about human crowd collision.

1 Introduction

There are very few studies on crowd modeling. We may mention the following related papers: C.
Reynolds¹developed a model for simulating a school of fish and a flock of birds using a particle

systems method ²; D. Terzopoulos³developed a model for behavioral animation of fish groups based on

the repertoire of behaviors which are dependent on their perception of the environment; S. Velastin⁴

worked on the characterization of crowd behavior in confined areas such as railway-stations and shopping
malls using image processing for the measure of the crowd motion; T. Calvert⁵developed the

blackboard architecture that allows the animator to work cooperatively with a family of knowledge based
tools; E. Bouvier⁶presented a crowd simulation in immersive space management and a new approach of

particle systems as a generic model for simulations of dynamic systems ⁷ .

In this paper we present a new approach of crowd behavior considering the relationship between groups

of individuals and the emergent behavior originated from it (i.e. the global effect generated by local

rules). We treat the individuals as autonomous virtual humans that react in presence of other individuals

and change their own parameters accordingly. In addition we describe a multiresolution collision method

specific for the crowd modeling.

This paper is structured as follows. In section 2, we present some sociological concepts of crowd

modeling which were used in our application. In section 3, we present information concerning the

model : individual and group parameters, distributed group behavior and implemented sociological

effects. In section 4 we present the scenarios where we have applied our model. In section 5, we describe

our collision avoidance methods and we present some results and analysis of this problem in the context