Screen space point sampled shadows

Abstract

It is no surprise that shadow mapping and its variations are the defacto standard for shadow generation in video games. Shadow maps are easy to compute in real-time and its many variants produce high quality shadows at interactive rates. With simple filtering techniques or by incorporating multiple shadow maps per area light source, soft-shadows can also be produced at interactive rates depending on the scene complexity. Despite these benefits, there are still a number of limitations inherent with shadow mapping-based techniques. More specifically, these algorithms are prone to sampling and aliasing problems, they cannot handle omnidirectional point light sources properly, and they require a shadow map to be rendered for each light source thus limiting the number of shadow-casting lights in the scene. In this paper, we propose a novel method for generating shadows that overcomes the issues inherent in traditional shadow mapping techniques. Our technique involves neither rendering the scene from the point of view of the light, nor expensive extrusions of the geometry. Instead, shadows are calculated in screen space using a series of depth textures that are shared by all of the lights, allowing the method to render a high number of shadow-casting light sources at real-time frame-rates on modern hardware.