Real-time GI for dynamic environments using temporal filtering

Abstract

Real-time rendering applications have traditionally shunned from implementing fully dynamic global illumination solutions due to their high computational cost, despite the added value in terms of realism and image quality the latter can afford. However, with the recent advent of rendering hardware accelerating ray tracing-based methods, new avenues of research have opened up that investigate the use of offline global illumination rendering techniques for real-time purposes. In this dissertation, we propose a method for interactive global illumination that reformulates IGI, a traditionally CPU-based rendering approach, and augments the derived solution with information from previously rendered frames. Theproposedmethodisfullydynamic,inthatnopre-computationisrequiredand scene objects such as light sources, geometry and materials may be updated without any performance penalty. We evaluate the method’s performance in both static and dynamic scenes, and compare the resulting image quality against ground truth images generated from an unbiased offline renderer. Results are promising and show that the method may be viably used for interactive applications, without degrading the quality of the output.