Calibração de fontes de luz pontuais baseada em sombras

• Ano de defesa: 2005
• Autor(a) principal: Vilar Fiuza da Camara Neto
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; UFMG
• Palavras-chave em Português: Visão computacional; Visão por computador; Computação; Robotica; Processamento de imagens \xTecnicas digitais
• Link de acesso: http://hdl.handle.net/1843/RVMR-6EAK5X

Most researches done so far about punctual light sources are based on the model of directional sources, i.e., they determine the direction of an illuminant placed in an infinite distance from the scene. This model is convenient mainly for solar-light based applications, but turns out to be inadequate when the illuminants are relatively close to the scene an indoor typical situation, where the illumination often arises from a set of lamps located a few feet from the observer. In this work we propose a novel approach for this problem. Its main advantage is to consider that the sources arent far from the scene. So, the objective is to retrievethe sources 3-D coordinates instead of its direction. The proposed method doesnt care about objects surface properties (the so-called BRDF), is based on very little knowledge about the scene, and is statistically optimal, because error metrics follows a realistic noise-propagation model. In parallel, the method is able to perform partial scene reconstruction: it retrieves the 3-D coordinates of certain points of objects, such as vertexes and tips. A series of tests were performed both with computer-generated simulations and realscenes. They prove that the method is precise even in presence of expressive noise.Besides, the average time spent to solve an instance of the problem about tenths of a second in a common PC suggests that the method can be used in real-time applications.