Crowd Trajectory Prediction And Behavior Simulation Based On Game Theory

Crowd simulation is one of the main research direction of computer graphics,which has a wide range of applications in security control,internal planning of the building,group accident deduction and avoidance,etc.Since the crowd movement is the result of multiple environmental factors,it is a difficult and complex task to predict and simulate crowd behavior in complex real world scenes.As a traditional subject in the field of economics,game theory has brought new research perspectives and ideas to explore effective methods of studying crowd behavior.For the above problems,this thesis focuses on the conflict behavior in the crowd,studies the internal game mechanism contained in different scenarios,and predicts and simulates complex group behavior.The main work and achievements of this thesis are as follows:In terms of crowd behavior prediction,this thesis takes a special scene from the perspective of unmanned vehicle movement as an example,and puts forward a crowd trajectory prediction method based on game theory that can be applied to most dynamic scenes.For crowd at crossroads,their psychological characteristics are analyzed.A zero-sum game model based on human-vehicle conflict is established,and the game mechanism in the process of human-vehicle conflict is elaborated.On this basis,a pedestrian path prediction network is proposed,which takes the pedestrian pose,position and camera parameters as the input of the convolutional neural network to predict the pedestrian path in the field of vision,so as to guide the vehicle to selfnavigation and avoid the recurrence of conflicts.The experimental results show that the pedestrian trajectory prediction method combining game theory and convolutional neural network can better predict the trajectories of pedestrians than the previous methods.In terms of crowd behavior simulation,this thesis puts forward a simulation model of crowd behavior based on game theory based on the background of fan riots,which simulates the situation of tripartite confrontation between the mobs of two opposing camps and the police.Firstly,the “Three Big” model is used to establish the personality characteristics of each agent,and the influence of external factors such as environment is considered comprehensively.Secondly,combined with the SIR model(Susceptible Infected Recovered Model),it reveals the generation and transmission mode of anger,and simulates the complete process of fan conflict from brewing to outbreak.Finally,in order to further optimize the agent strategy,genetic algorithm is introduced.The comparison between the simulation results and the real situation shows that the model can well simulate the complex dynamic changes of fan conflicts,analyze the basic properties of similar riots,and provide valuable reference measures to avoid the occurrence of riots.