Research Of Multi-level Virtual Crowds Simulation Technology

We can see crowds everywhere in our lives, such as birds moving south, fishesin rivers, crowds of human beings. The technology of crowd simulation is widelyapplied for real-time simulations, virtual entertainments and some other fields,which has gradually become a research hotspot of the field of virtual reality inrecent years. The main purposes of simulating crowds are to analyze theself-organization phenomenon of communities, and to increase the dynamic realismfor static scenes. The states of the individuals in the crowds change dynamicallywhich is provided with a certain degree of intelligence and there is complexrelationship among the individuals, so it is a complex problem to model crowds. Atthe same time crowd simulation involves wide range of fields, such as computergraphics, robotics and so on, which greatly increases the difficulty of realizing thesimulation.Crowd simulations can be of vary contents in different application areas, butthere are three common problems to be solved, which are the modeling of crowds,path planning and collision avoidance. We solve the three issues in complexscenarios, which is complex in both the structure of the crowds and theenvironments. The individuals in the crowds are of different velocity, differentspeed, different social roles and different goals, even the environments consist ofstatic obstacles and dynamic obstacles. In order to simplify the group structure, themodel proposed in the paper is based on the ViCrowd framework. The whole crowdis divided into three levels, which are individuals, groups and crowds. Therelationship of the individuals in the same group is defined as a dependency matrix.The3D model of the human and the environments are constructed by Creator asoftware developed by MultiGen. First parse the scene to get the two dimensionalboundaries of the static obstacles by OpenFlight API. Then treat the obstacles to beprimitives and generate the generalized Voronoi diagram, and figure out the weightof the edges of GVD, finally export the diagram as a global navigation map forAgents. Since the edges of two adjacent primitives point in the vertical position ofthe line linking the two primitives, using GVD as a global navigation map we cancalculate the shortest route for Agents, also Agents will walk away from theobstacles if walking along the edges. We propose an approach based on priorityrules to deal with collision avoidance problem among the Agents. Firstly, wecalculate the individuals in the view of the Agent with a synthetic visual model.Secondly, predict whether the individuals within the field of view will collide with the Agent, and collect the individuals who will collide with the Agent in the nearfuture as a set. Thirdly, divide the set into five parts according to the collision type,and partition them into four priority levels. The predicted collisions to be happenedimmediately should be tackled first. Fine-tune the velocity of the Agent based onVO model in the case that no collision is going to happen in the near future,therefore eliminate the possibility of collision at an earlier time.The individuals are organized as groups, so the module handling the groupevents is included in the model. We can observe the phenomena in the crowdsimulation experiments that Agents walk without colliding with others in therouting process and groups split in collision avoidance process and Agents regroupafter the elimination of collision.Finally, four groups of simulation experiments are designed for trafficintersections and shopping malls, which verify the approaches proposed in the paper.Meanwhile, the paper gives a quantitative description of the frame rate andreal-time simulation results. Experimental results show that crowds modeled as setsof groups are more realism and efficient than that composed of individuals.