Research On Heterogeneous Pedestrian Motion Simulation Based On Velocity Field And Social Force

The security incidents from crowd activities in large-scale gatherings in public places have drawn a lot of attention from the society.It is an urgent problem to predict and reasonably control the abnormal behavior of the crowd in a complex environment.Due to huge resources from organizing a large number of personnel to conduct security drills,simulate crowd activities with the emerging virtual crowd simulation technology can effectively reduce costs.No matter from the economic benefits generated or on the level of maintaining social security,virtual crowd simulation technology has a strong demand and great significance.The existing mainstream solution for virtual crowd simulation is the agent-based model.By simulating individual pedestrians with different agents,the virtual pedestrian can perceive the surrounding environment and make corresponding behavior decisions based on the perception information.In order to build the virtual pedestrian motion decision model,in this thesis,the pedestrian motion decision model is divided into two parts:motion path planning and motion local collision avoidance based on the existing research.In this thesis,the data-driven velocity field method performs vertical path navigation for virtual pedestrians.After extracting pedestrian trajectories from real video data,this thesis classifies pedestrian trajectories gathered at different exits in clustering methods.The velocity fields of exits are built to guide the pathfinding process of the crowd in the scenario of multiple entrances and exits,which provides a distributed and reliable path from the entrance to the exit for the virtual crowd.At the same time,in order to simulate the diversity of crowd pathfinding,this thesis use randomization to improve the velocity field model and the experimental results show that the pedestrian trajectory driven by the improved velocity field here can well simulate the real pedestrian trajectory.Considering that in reality each individual has its own unique behavioral characteristics,these different individuals constantly interact in the movement and together form a crowd with heterogeneous characteristics.This thesis divides pedestrian emotional infection and stress into considerations of social force models in the local collision avoidance of the virtual crowd,and divides the virtual crowd according to the improved Durupinar emotion model,reflecting the heterogeneous characteristics of the pedestrian crowd.Considering the crossing behavior of individuals in heterogeneous crowds,this thesis introduces the shift and side overtaking social force model under different crowd densities to simulate the movement process of pedestrians overtaking the individual in front,and designs experiments to verify the effectiveness of the improved social force model in collision avoidance and overtaking behavior.Experimental simulation results show that the approach in this thesis can achieve the heterogeneity of the crowd,and can meet the performance requirements of the system while reproducing the movement process,of the crowd in a given scene.