Eficiência e precisão no cálculo de iluminação por radiosidade com GPUs

Abstract

This work consists of a study about the global illumination model known as radiosity, in an effort to achieve an efficient implementation on modern hardware.Radiosity is a classic and very important method in computer graphics. It is the first proposed method to achieve photo-realism as it considers interactions among the objects of the scene. As the method has a high processing demand, we propose the use of modern GPUs, with their massive processing capacity, in an attempt to speed up the computation.We experimented with two distinct implementations that use GPUs to accelerate part or all of the method. The first implementaion is similar to the original proposal of radiosity, and the GPUs were used to solve the simultaneous linear equations. The second implementation is derived from an extension to the original radiosity method that was designed for processing in GPUs.As results, we compared different methods to solve the simultaneous linear equations, on both GPUs and CPUs and we proposed an improvement on the GPU implementation related to the determination of visibility. Our modifications improves the quality of the output image while preserving the interactive charateristic of the method variation.