基于样本的大规模人群快速创作

Fast and Large-Scale Crowd Authoring Based on Samples

人群作为虚拟环境中必不可少的一部分,能够极大地提高场景的真实感.文中提出一种基于样本的大规模人群快速创作方法,将人视为具有分布和角色信息的粒子,通过从真实图片中获取包含小规模人群粒子信息的样本,由样本建立Wang Tile集合,使用Wang Tile非周期性拼接覆盖场景区域,从而快速生成任意规模人群.文中的创作方法简单快速,用户只需输入若干样本和场景区域.同时,用户可以通过调整Tile的边颜色快速地进行局部人群修改,也可以使用密度图得到用户需要的效果.实验给出了3个场景的人群创作效果,表明文中的方法可以十分迅速地生成大规模具有真实分布信息的多样人群.

Crowd is an indispensible part of virtual environments （VEs）. Populating VEs with crowd can greatly enhance the reality of VEs. We propose a massive crowd authoring system based on crowd samples. Each agent is viewed as a particle with its distribution and character information. By utilizing the samples extracted from real crowd pictures that include information of small crowd particles, we build up the Wang Tiles set. By non-periodically tiling the plane of VE with the set of Wang Tiles, we can quickly populate the VE with crowd particles in an arbitrary size. Our method can create massive crowds in a simple and fast way. It only requires the user to input the samples and crowd region in the VE. The user can further adjust the crowd in local regions by simply changing the color of the Tile edges. The approach can also control the final crowd result by using a density map. We build up three scenes and populate them with Wang Tiles that contain crowd particles. The results show that we can create massive and diverse crowds with realistic distributions in a very fast way.