Fleeing Pattern Simulation in Emergency Situations

Abstract

Large crowd gatherings are a common and frequent experience in modern society. Predicting how large crowds will react in certain situations such as panic situation has long been under investigation especially to overcome crowd disasters. How can we evaluate the evacuation e ciency or predict how crowd will react to a certain emergency situation? Most direct approach is to conduct a real time evacuation drill which is both time consuming and expensive. Further these drills are considered dangerous and will not re ect the real scenario since psychological in uence is di erent from a real emergency. Computer based simulation tools provide an interesting alternate for the problem. These tools eliminate the need of real participants thus more cost e cient. Also can perform repeated tests as required and can have many features enabled such as vary the no of agents and change the oor plan. This thesis describes a research for eeing pattern simulation in emergency situations which integrates both swarm intelligence and fuzzy logic parameters in a goal nding multi-agent Environment. The proposed model uses a modi ed particle swarm optimization algorithm for path planning which produces dynamic and non-deterministic paths for agents, towards the goal while avoiding collisions with static obstacles and other agents. By nature human behavior is unpredictable due to human behavior characteristics. Many factors, such as a person's emotions or intelligence, can determine how they act when they are threatened with an emergency making the prediction more challenging. In this research fuzzy parameters like stress and panic are incorporated for re ning human emotional behavior which brings a sense of reality for the simulation. A prototype system has been developed in C++ which is able to demonstrate several crowd behaviors such as herding, leader following and clogging in a simple 3D indoor environment. Ogre 3D graphic engine has been used for building the visualizer. The evaluation of the model was done through a performance evaluation since conducting a user evaluation is too expensive and di cult.