Parametrização da iluminação de uma cena virtual 3D baseado em imagem LDR

Abstract

The popularization of Graphic engines is constantly growing, allowing more users to use 3D virtual scenarios in their projects, without much knowledge of computer graphics. However, there is a limit to the realism that can be achieved without the development of advanced rendering techniques, especially when it comes to lighting the scene. Techniques that intend to increase the quality of illumination are a recurring theme of study in the area of computer graphics. In particular, image-based lighting is a technique widely used for its realistic results, but it is limited by the use of images with high dynamic range, which are usually only captured through special hardware. Considering this, this work presents a solution for automatic conFiguration of lighting parameters inside the Unity graphics engine using as input only a simple outdoor image, in order to make the virtual scene have the same light conditions as the input image. For this, was implemented a neural network architecture that was trained to estimate lighting information from the input image. This network is then imported into a program where the estimated values are converted into parameters and applied within Unity’s lighting algorithms. The technique is capable of using photographs of different weather situations to configure multiple parameters in a fully automatic way, innovating the way to configure the lighting of a scene. The tests prove the high quality of the results and success in achieving the objectives.