Modelagem automática de cenas com iluminação local a partir de imagens

Abstract

This thesis deals with the problem of acquiring the models of a scene from its images, where the correspondence between the image pixels intensities and the real colors of the scene cannot be determined using a local illumination model of the interaction of the light with the scene. More specifically, the problem is treated in two particular cases in distinctscenes.In the first case, we have the underwater scenes in muddy water where the light interacts globally with the environment, to which we developed a methodology to extract the scenes geometry, by integrating a light propagation model into a stereo vision algorithm and recoveringthe disparity maps from images pairs acquired from distinct viewpoints. Experimental results are compared to known methods. The method allows to restore the images of theunderwater scene.In the second case, we have a scene where the light interacts globally with objects that has homogeneous, reflexives and translucent surfaces, presenting shades and Inter-reflection between objects, to which we developed a method of recovering the reflectance parameters to each object of the scene, by solving a factorized equation system obtained using just one rendering and having as restrictions an image of the scene, whose solution is more steady and efficient in comparison to similar methods.