Impostor-based crowd rendering

Real-time rendering of detailed animated characters in crowd simulations is still a challenging problem in computer graphics. State-of-the-art approaches can render up to several thousand agents by consuming most of the graphics processing unit (GPU) resources, leav- ing little room for other GPU uses such as driving the crowd simulation. Polygonal meshes deformed through skinning in real time are suitable for simulations involving a relatively small number of agents, since the rendering cost of each animated character is proportional to the complexity of its polygonal representation. A number of techniques have been proposed to accelerate the rendering of animated characters. Besides view frustum and occlusion culling techniques, related work has focused mainly on providing level-of-detail (LoD) representations. Unfortunately, most surface simplification methods do not work well with dynamic articulated meshes. As a consequence, the simplified versions of each character are often created manually and they still suffer from a substantial loss of detail. Image-based precomputed impostors for the whole character provide substantial speed improvements by rendering distant characters as a textured polygon, but suffer from two major limitations: all animation cycles have to be known in advance (and thus animation blending is not supported) and resulting textures are huge (as for each view angle and animation frame, an image has to be stored). In this chapter, we overview different approaches to crowd rendering, focusing on impostor-based techniques. We summarize and compare two recent approaches [1,2] based on rigidly animated impostors per body limb. Compared to having impostors representing an entire character, having animated per-joint impostors provides a more memory-efficient approach.Peer ReviewedPostprint (published version