Data-Driven Macroscopic Crowd Animation Synthesis Method using Velocity Fields

This paper address the problem of synthesizing realistic crowd behavior animation in virtual environment, which usually requires setting a lot of empirical parameters and interacting with the users frequently. A new method is proposed in this paper, which extracts the typical pedestrian velocity fields from video data, and then synthesizes the macroscopic crowd behavior animation based on these fields. The video data used to extract velocity fields was generated from a specifically designed experiment. With these velocity fields, we could design the macroscopic crowd behavior interactively in a 2D behavior editor, and synthesized the corresponding real time 3D crowd animation. Experiments results have proved that this new method can get a good performance of realistic crowd animation synthesis with little user interaction, and can be applied in many ways.