Introduction and Motivation of the Thesis

Realistic animation has been an important goal in Image Synthesis since the beginning of Computer Graphics. An increasing number of applications is emerging continuously following advances in software and hardware graphics. Applications of realistic animation can range in a wide spectrum from classical ones involving CAD systems (architecture and engineering design and prototyping, or flying simulations), to new ones from the emerging fields of virtual environments and augmented reality. Film productions of 3D realistic animations and games also present an important improvement in the last years as a result of these advances.

One of the most important topics involving the generation of realism in an animated sequence is the accurate computation of the illumination. This topic has focused the attention of many researchers in the last twenty years. Physically-based global illumination models were successfully developed to simulate the process of interaction between the light and the environment. Ray-tracing, Radiosity and Hybrid methods are currently used for realistic rendering in many applications. However, they are still time-consuming methods if they are used for the generation of complex scenes involving complex light transport phenomena. For animations, generally composed by sequences of lots of images, this problem grows, then leading to large amounts of computation if each frame is treated independently. So, efficient solutions are searched in order to reduce the computation cost.

Considering that changes in illumination between frames are usually smooth as a consequence that animation changes are also smooth, a natural possibility for an efficient approach is to improve global illumination models in order to make use of this source of coherence. For example, a significant computation time can be saved by reusing illumination already computed that is invariant in an interval of time within the animation.

The radiosity method, proposed in, allows interactive walkthroughs since it is a view-independent global illumination solution. It is an attractive method to be used both in animations and in interactive applications. Many works with a focus on this last domain of applications were proposed. Strategies like progressive refinements methods or hierarchical radiosity have been developed with the goal of quickly update changes in illumination for a given demand. In the case of non-interactive radiosity animations, where all time-dependent parameters can be known a priori, requirements are not the same as in the interactive case: a high number of a-priori-known frames must be computed maintaining a minimum level of accuracy, thus, emphasizing more in quality than in speeding updates of single frames. Although in most cases interactive techniques could be adapted for this purpose, a specific approach for animations can extract more benefits.

The objective of this thesis is the development and implementation of new methods to compute animations in radiosity environments. The two primary goals of our approach pursues are efficiency and correctness. The computation of a globally-illuminated animation is an expensive task, we will therefore focus the efficiency and optimizations that can reduce the computation time. The lighting simulation resulted must be reliable, we base our method in a physical model of the global illumination problem using Monte Carlo methods.