New approaches to real-time global illumination algorithms for dynamic scenes

In this thesis new approaches to real-time global illumination algorithms for dynamic scenes are proposed. Methods on this topic proposed in the literature are investigated and three new methods that complement each other are proposed. The first proposed method is based on importance sampling of environment maps. Environment map images, parametrized in spherical coordinates as a two-dimensional piecewise function is divided into blocks using a kd-tree structure. By representing the probabilities of these block with a simple model, importance sampling of environment maps can be achieved. The second method aims to handle visibility tests in real time. For this purpose, objects in the scene are decomposed into convex parts which are then stored in a spatial data structure. Convex parts that needs to be tested can be determined by traversing this data structure. Finally, each convex part can be tested with the proposed real-time convex - ray intersection test. The last method aims to approximate rendering equation in real time. For this purpose, Bidirectional reflectance distribution function (BRDF) is modeled by a mixture of spherical normal distributions which simplifies the rendering equation.