Visibility Culling

One of the most important problems in computer graphics is the classification of all objects in a scene as either being at least partially visible or totally invisible. The issue of rapidly identifying entirely invisible objects is of importance in any rendering system, but it is especially crucial when interactive frame update rates are desired. In an ideal rendering system it would be possible to exactly identify those objects that do not contribute to the generated image in any way at zero cost and, furthermore, to draw only the visible parts of those objects. Since we do not live in an ideal world, however, we try to make the process of determining all objects that are visible from an arbitrary dynamic viewpoint in a scene as efficient as possible. Thus, we want to cull all objects of a scene which are entirely occluded by other objects, or not even contained in the viewing frustum, in large chunks and with minimum cost per culling operation. This paper first gives an overview of several issues pertaining to virtually all visibility culling algorithms and then examines some of the most important work in detail. Additionally, we try to assess how some of these algorithms depend on the availability of a hardware z-buffer and how, if at all, they are practically applicable to software-only systems.