Research On Virtual Pedestrian And Crowd Simulation Based On Multi-agents

Virtual crowd simulation is an important research direction in the field of computer simulation.There are strong practical needs in various crowd simulation fields such as digital entertainment,public environment planning,and disaster prevention simulation about large-scale virtual crowds dynamic walking.How to quickly construct a pedestrian simulation model,and to achieve the real-time and even ultra-real-time computational efficiency,to achieve high quality simulation effect is a hot and difficult problem in current crowd simulation research.An agent is a computing entity that can sense the environment and simulate intelligent behavior.It was difficult to apply to large-scale crowd simulation due to the limitations of hardware and software resources.With the breakthrough of artificial intelligence algorithms and the substantial improvement of computer hardware level,the use of agents for large-scale crowd simulation has gradually become an im portant research method.In the research of this thesis,the problem of autonomous navigation of the agent and the microscopic perspective for multi-agents is considered separately,to achieve high quality crowd simulation in the large-scale scene.Autonomous navigation focuses on how to make the agent perceive the environment information and realize fast path finding.The micro-interaction mainly studies the collision avoidance problem between the agent an d other agents or obstacles.In the research of agent autonomous navigation,considering the requirements of authenticity and computational efficiency,this thesis uses symbolic regression to distill behavior rules from trajectory of real pedestrians which extracted from video data.In the process of solving the specific symbol regression problem,this thesis optimizes the implementation based on SL-GEP algorithm to obtain faster convergence speed.The experimental results show that the proposed method can guide virtual pedestrians to quickly generate walking trajectories similar to real pedestrians and also has good performance.In the research of multi micro-interaction of multi agents,in order to improve the authenticity of virtual pedestrian simulation local dynamic collision avoidance,this thesis introduces the concept of pedestrian instantaneous energy consumption based on the classical underlying ORCA algorithm,and designs the optimal energy consumption.Local dynamic obstacle avoidance algorithm.In addition,this paper uses the telescopic collision box method based on ellipse model to simulate pedestrian side collision avoidance behavior.The experimental results demonstrate the effectiveness and efficiency of the proposed method.The work of this thesis has obtained a better simulation effect of virtual crowd,and it can also meet the needs of real-time simulation tasks in terms of computational performance.At the same time,the method of this thesis has better scalability,and it has certain reference and application value for multi-role and personalized crowd simulation.