Massive Crowd Simulation In The Complex Environment

The massive crowd simulation in the complex environment plays a significant role in the area of urban planning,early warning of disaster,video game and movies VFX.The accurate crowd simulation aids people to evacuate in the catastrophe,which reduces casualties.Meanwhile,it provides the life-like interaction to the scene for players and audiences in the video game and movies.The traditional methods use mathematical way to calculate the path and the collision avoidance in the crowd simulation.Such as Dijkstra algorithm and A* algorithm for pathplanning,and the RVO,RVO2 library for the collision avoidance.The algorithms offer optimal solution mathematically,but they failed to simulate the real world more or less.This thesis aims to solve the problems against such background.It proposes to calculate the collision avoidance by using the neural network and do path-planning with the human emotion emulation.The following shows the main research of the thesis.1.To train the neural network,real trajectories are required as the input.The thesis combines ASMS video tracker with the depth prediction network to improve the method of extracting of pedestrian trajectories from real videos.Therefore,the tool to extract 3D trajectories from video shot by the single-camera is created.2.Path calculated by mathematical equation is accurate.In the real world human beings do not always find the optimal routes for themselves.To simulate the decision of human beings,the improved path-planning method with the emotion emulation is proposed in the thesis.The stimulus and the personal intrinsic quality are used in the emotion emulation equation to calculate the human emotion in the method.Then the emotion will be mapped as a factor to adjust the path.3.The thesis improves the existing method to simulate human motion based on the real data.It proposed to use the real data to train the neural network in order to simulate human motion.The method is expected to improve the realness of the simulation.Utilizing the improved methods,the system to simulate the massive crowd is implemented in the thesis.The system is able to extract pedestrian trajectories in the 3D world from the real videos,and to simulate the massive crowd behavior,etc.The test to the system shows that it can obtain the life-like crowd simulation.