A Hybrid GPU Rendering Pipeline for Alias-Free Hard Shadows
Stefan Hertel, Kai Hormann*, Rüdiger Westermann
Computer Graphics and Visualization Group, Technische Universität München, Germany
*Technische Universität Clausthal
Background
We present a new GPU pipeline for rendering per-pixel exact shadows that are cast by point lights and parallel lights. Our approach is hybrid in that it uses kD-tree accelerated ray-tracing to determine shadow-ray intersections, and rasterization to effectively reduce both the number of shadow rays to be traversed and the number of sub-spaces to be considered along each of these rays. To achieve this we introduce conservative shadow maps, which store a conservative estimate of the first intersection with the scene for each possible shadow ray. A novel approach to build such a map is presented, which uses rasterization to compute for every shadow-map pixel the triangles intersecting this pixel. By exploiting the rasterization capacities of recent GPUs in combination with accurate ray-triangle intersection tests, we are able to efficiently compute alias-free shadows in high-resolution and spatially extended scenes where classical shadow mapping techniques have severe difficulties.