Voxel Based Real-Time Global Illumination Techniques
Student: Daniel Gomes
Supervisor: António Ramires
Abstract
One of the goals in computer graphics is to be able to generate photo-realistic images and do it in real time. Unfortunately most lighting algorithms are not able to satisfy both objectives at the same time.
Most of the algorithms nowadays are based on rasterization to generate images in real time at the expense of realism, or based on ray tracing, achieving photo-realistic results but lacking performance, which makes them unsuitable to compute at interactive frame rates with the computational power currently available.
Over the last years, some hybrid approaches have emerged that try to combine the best features of both types of algorithms.
What is proposed in this thesis is the study and analysis of a class of algorithms based on voxels to approximate global illumination in 3D scenes at interactive frame rates.
The goal of this study is an analysis on the practicability of such algorithms in real-time applications. It is shown that these techniques are very similar in their structure. They use a volumetric pre-filtered representation of the scene and a rendering algorithm based on ray tracing or cone tracing to compute an approximation to global illumination in real time.
The main problem of filtering the appearance of objects in the scene using voxels is that these representations can require quite a large quantity of memory. Most of the described algorithms try to deal with this problem in some way and so multiple data structures are presented to try to solve the problem. Another approach is described that uses the wasted space of a regular data structure to compute volumetric effects together with the approximation to global illumination, in real time.
A selection of the described algorithms are implemented and analysed in terms of performance and visual quality of the resulting image and some conclusions are drawn, describing some possible future improvements.
Thesis Download (PDF)