Agent-based Modeling And Simulating For Crowd Behaviors

With the development of society and the growth of population,gathering of large-scale crowd has become a common social phenomenon.Due to the lack of guidance and control of the behavior of the crowd,there have been a number of incidents and disasters,which have caused great harm to public safety.Agent-based crowd simulation can simulate and predict the possible crowd behavior in advance,find out the problems and take corresponding preventive measures.The crowd simulation can be used to analyze the crowd dynamics both qualitatively and quantitatively.In order to solve these problems,various behaviors of the crowd need to be reproduced and modelled.This requires an in-depth analysis of the underlying patterns and principles that affect the human behaviors.In this thesis,the existing work on crowd simulation is surveyed and analyzed.In order to construct a realistic and effective crowd simulation system,we focus on the crowd behavior modeling from two aspects: the crowd movement behavior at the bottom level and crowd decision making behavior at the top level.The movement behavior modelling refers to individuals' avoidance of obstacles and other people by adjusting the direction and speed of movement.The decision making behavior modelling refers to the decision choice that an individual makes,such as the selection of the evacuation exit.The crowd movement behavior is affected the decisions made by the individuals in the crowd.These two types of behaviors form the basis for simulating crowd dynamics.Our contributions in this research are as follows:(1)For crowd movement behavior modelling,a proactive steering behavior model is proposed,which is capable of simulating the crowd behaviors such as gap seeking and following.The proposed model is integrated into a layered crowd simulation framework,which also includes path planning and reactive collision avoidance module for simulating the crowd dynamics.(2)For crowd movement behavior evaluation,the evaluation methods including 3D visualization,fundamental diagrams,trajectory similarity and progressive distance error are used to compare the proposed model with classical crowd models.The simulation results demonstrate the advantages of our model over other models from both macroscopic and microscopic perspectives.The advantages of the fuzzy rule-based model has been shown by comparing with the traditional discrete choice model.(3)For crowd decision making behavior modelling,a fuzzy rule-based representation is used for constructing the decision rules of individuals in a crowd.A genetic programming approach is adopted to identify the decision rules from crowd data.(4)With the use of simulation and real data,the effectiveness of the identified decisions rules are validated.