Interactive Control Of Large Crowd Navigation

With the rapid development of computer graphics technology and hardware, computer animation has wide application in areas such as motion picture, entertainment, advertisement, architecture, engineering, art, education, computer games, etc. Thanks to the prosperous economy and living standard, novel visual effect and vivid picture produced by crowd animation techniques have occupied mass media and become an important part of everyday life.Crowd is composed of intelligent agents, which independently determine their actions while obey basic rules. The emerging behaviors of these individuals reflect the behaviors of whole crowd. Thus, motion control has become the most important and difficult part of simulating vivid crowd. Nowadays, how to meet the industry needs of interactive control and real-time rendering of large crowds is the key to the development of crowd animation.In this thesis, we try to reproduce large crowd virtual agents and provide users with simple and flexible control tools. Our simulation system focuses on controlling the movement of crowd. Our controlling strategy allows real-time user control and simulation rate.We approach the crowd control research area from two aspects. In the first approach, we investigate the flock behavior in 3D space with user defined dynamic shape constraints. In the second approach, we propose a hybrid framework mainly to animate pedestrians in urban area from a global viewpoint.In Chapter 2, we propose a shape constrained flock animation framework. First of all, every flock member complies with three predefined basic rules, namely separation, cohesion, and alignment rules. At the same time, attracted by virtual shape constrained points, an individual moves smoothly along interatively defined trajectory towards the destination. During simulation, with the help of Kalman dynamic tracking scheme, the whole flock will behave according to a series of shape constraints. Our system can produce interesting visual effects such as flock moving in dynamic or static shapes.In Chapter 3 and 4, we employ a global control theme to analyze the behavior of large crowd of pedestrians in real-time. Once a velocity field covering the whole simulating area is created, everyone will be driven by it. During the simulation, users can modify the velocity field interactively, thus change the motion of pedestrians. If a crowd is composed of several groups, we can use one velocity field to control one group. We propose two methods to calculate the velocity field. One way is to specify velocities in some scattered points, then get the desired velocity vector field through Radial Basis Function interpolation. This method can create a smooth vector field in a fast and simple way. This method needs a relative small memory space and the field accuracy is independent of the grid resolution. The other way is to take a pedestrian flow as a water flow. By solving the Navier-Stokes equation which describes 2D incompressible fluid, we obtain a velocity field. The advantage is that we can tune the velocity field by adjusting the parameters of the Navier-Stokes equation. Because the Navier-Stokes model is based on the fluid dynamics, it is approprite to exhibit results such as several groups of pedestrians join together, pedestrians wriggle their way.To render thousands of agents, we employ a LOD hierarchical model to balance between computational time and rendering quality. We switch between detailed mesh models and Impostor images according to the distance between agents and the viewpoint.