Research On Crowd Behavior Simulation Based On Hybrid Data-and Model-driven

With the application field of simulation technology becoming more extensive,more and more scholars pay attention to the research of virtual crowd behavior simulation.The application of simulation technology of virtual crowd can be seen from people's daily life and entertainment to military exercises of troops.Yet there exist such issues as less authenticity of the models and virtual human behavior unable to truly reflect the state of human movement in the current simulation technology of virtual crowd.Aimed at these problems,this paper develops deep researches on the vital technology of virtual crowd behavior simulation.Virtual crowd behavior is discussed and researched respectively from the framework design of crowd behavior simulation,method study of data-driven simulation modeling,affective model of virtual human as well as the path planning algorithm and other aspects.The corresponding algorithm and model are proposed.On this basis,the simulation demonstration system of virtual crowd behavior is established based on the hybrid data-and model-driven methods.The main contents of this paper are summarized as follows:(1)First,the research background and significance of the topic are elaborated in the paper.And the research results of this research group in the field of virtual human for years are sort out.Then the research status and development trend at home and abroad are summarized.The overall framework and research contents of the paper are concisely discussed.(2)Second,aiming at the issue of framework design of crowd behavior simulation,with the combination of data-driven and model-driven methods,an efficient simulation modeling method of crowd behavior is proposed based on the hybrid data-and model-driven methods,and simulation framework of crowd behavior is established based on the hybrid-drive in this paper.This framework can not only use the collected data of real crowd behavior for correction model-driven,so as to establish a more lifelike crowd behavior model and greatly reduce the workload,but also improve t he adaptability and scalability of simulation framework of crowd behavior.(3)Third,aiming at the data-driven method,optical flow algorithm is used to extract feature information of crowd behavior in this paper.Video information of the crowds is first gathered for image processing at the square of railway station,and then the processed image generates the vector field of crowds through optical flow algorithm.(4)Fourth,the affective modeling research of virtual human is carried out by Hidden Markov Model.Some riots occurred recently are firstly collected and analyzed in this paper.The corresponding contact between individual behavior and facial expressions is established.Then through the establishment of three-dimensional affective space Hidden Markov Model,the experimental analysis of facial expression and mood swings is made.Finally,the relationship between the individual inner moods and behavior is established.(5)Fifth,in-depth research of path planning algorithm of individual is carried out in this paper,particularly the algorithms of Probabilistic Roadmap Method(PRM)and social force model.Theoretical analysis and experimental research are firstly made for the two algorithms.And aiming at such problems as too few path milestones generated in narrow passages and boundary regions,too redundant generation path by Probabilistic Roadmap Method and the individuals on the opposite collision frequently in social force model,the corresponding optimization plan is proposed.Finally,the effectiveness of optimization plan of the two algorithms is verified experimentally.(6)Sixth,the simulation demonstration system of virtual crowd behavior based upon the combination drive of data and model is finally developed in the paper.Microsoft Visual Studio 2013 and QT5.4 are used to develop the simulation demonstration system of virtual crowd behavior with good visibility and interaction.With the scenario of group events combined,simulation experiment analysis of crowd behavior is made.The feasibility and effectiveness of relevant models and algorithms in paper are further verified.Some guidance can be offered for group events in the real world.