Rasterization + Ray Tracing: Rendering of Hard Shadows

Student: Fábio Morais
Supervisor: António Ramires

Abstract

Due to great technological advances in video cards over the last decade, several classical imagerendering algorithms have recently been adapted to run on GPUs. This made it possible for several ray tracing based global illumination techniques to perform faster and faster, achieving performance levels which are, in some cases, suitable for real-time applications. However, despite these advances, rasterization is still the most widely used technique in the 3D Computer Graphics industry for real time applications due to its efficiency generating images with reasonable visual quality. As the implementation of photorealistic techniques using ray tracing in real-time is still out
of reach of today’s hardware, there have been several attempts to combine rasterization and ray tracing, to obtain the best of both worlds.

This dissertation seeks to explore the benefits of an approach that combines rasterization and ray tracing and to study the feasibility of this approach using as case study the creation of hard shadows. There have been two relevant works on this particular area in the past. The first goal of these methods was to identify problematic pixels in a shadow map, i.e. pixels which according to the method could not be determined with certainty to be in shadow or light. One of such methods guarantees the correctness of the remaining pixels, whereas the other is based on a heuristic approach. The set of pixels in doubt was then sent to a ray tracer to evaluate their shadow status.

In this work we also explore an extension to one of such methods that takes into account information regarding the triangle’s adjacency, thereby being able to guarantee the shadow status of more pixels based only on the data provided by the shadow map. To assert the usefulness of such approach we evaluate all three methods against a ray tracing solution, analysing the graphics pipeline, the overhead in terms of computational effort on the rasterization front, the number of pixels sent to the ray tracer, and also the visual impact of the pixels that are incorrectly assumed to be accurately classified by the shadow map.

Thesis Download (PDF)