Research And Implementation Of Data-driven Based Crowd Simulation Method

Crowd simulation is applied to fields of science,commerce,entertainment and so on,and has attracted extensive attention in recent years.The purpose of crowd simulation is to model crowd and simulate human behaviors in crowd.Since crowd is a sophisticated self-organized system,there are various factors influencing pedestrian motion and comprehensive applications need crowd simulation.So there exist many challenges when implement high-quality crowd simulation results,which also embodies the value of this work.This work concerns three levels of crowd simulation,including global path planning,local collision avoidance and body animation realization.To begin with,apply Delaunay triangulation into scene modeling,attain the static feasible paths of the scene and then use Dijkstra algorithm to calculate the shortest path from origins to end points of pedestrians.Because there do not just exist static obstacles,avoiding collision with surrounding people should also be considered.In order to achieve real and efficient collision avoidance result,this work mainly does research in the data-driven based simulation method.For the method of local collision avoidance,this work makes use of human-movement trajectories to build example databases;During simulation,virtual figures search similar examples in the databases according to their states;By means of collision prediction,actions that will not collide with other virtual figures or obstacles are selected from the similar examples,and make the virtual figures copy the actions.However,the method depends on pedestrian motion data.When lacking data,collision is easy to happen and meanwhile in case of large volume of data,there exists the problem of efficiency decreasing due to increasing data searches.Dealing with the two problems,rules of collision-free velocity calculation and a collision checking and eliminating algorithm are introduced to make collision seldom occur even when the data do not cover the simulation;exploiting motion trajectory data to train some artificial neural network and utilizing it to predict the simulated individuals' behavior can get rid of dependency on data during simulation and make the efficiency not be affected by the amount of data.Finally,make use of motion capture data and the data visualization method to generate the body movement.Extending pedestrian movement trajectories to crowd walking animation makes the simulation more real and complete.Simulation result shows that this work is able to model the scene,perform global path planning,make great improvement in collision avoidance simulation,and add body animation to the motion trajectories.Moreover,the prototype of the crowd simulation system is developed,integrating the global path planning,local collision avoidance and body animation.The system prototype realizes the graphic user interface which helps users or researchers understand the crowd simulation system more conveniently,watch the results of every simulation phase intuitively and facilitate future research work.