Visibility acceleration using efficient ray classification

Evaluating the visibility between two points is a common task in global illumination algorithms. Traditionally, a ray is traversed through a hierarchical data structure that is able to cull primitives which are unlikely to be hit by the ray. However, a great amount of time is spent in traversing the acceleration structure when a scene contains highly detailed objects or complex geometry like hair or foliage. While these objects contain a lot of disconnected geometry, their visibility is often very regular. In this paper, we present an algorithm which is able to exploit this regularity. In a preprocessing step we create a data structure for each object which classifies large parts of the objects ray space as either blocked or uninhibited by the object. In a rendering phase, large amounts of intersection tests can be avoided by performing a fast look-up in our data structure. We show that our method can be advantageous in scenes containing complex geometrical objects since our algorithm is able to classify a large amount of shadow rays, avoiding expensive intersection tests using a traditional acceleration structure.nrpages: 8status: publishe