%0 Conference Proceedings
%4 dpi.inpe.br/banon/1999/01.13.17.58
%2 sid.inpe.br/banon/2001/03.30.15.53.51
%T Geração Automática de Mapas de Disparidade em Visão Estéreo
%D 1998
%A Nogueira, Fernando Marques de Almeida,
%A Tozzi, Clésio Luis,
%E Costa, L. da F,
%E Camara, G.,
%B International Symposium on Computer Graphics, Image Processing and Vision, 11 (SIBGRAPI)
%C Rio de Janeiro, RJ, Brazil
%8 20-23 Oct. 1998
%I Sociedade Brasileira de Computação
%J Porto Alegre
%S Proceedings
%1 SBC - Sociedade Brasileira de Computação
%K shape from stereo, disparity, matching, gradient
%X Tridimensional information of a scene can be obtained using the technique named "shape from stereo". The major difficulty of this technique is the automatic determination of correspondence between images (matching points). The stereo matching solution can be obtained from area-based matching and/or feature-based matching and represented as a disparity map. Most of the contemporaneous algorithms found in the literature employ feature-based matching techniques for the stereo matching. This class of matching produces a sparse disparity map, which is further interpolated in a smooth dense disparity map that does not represent correctly the scene abrupt variations in depth. An approach for stereo matching that results in dense and non-smooth disparity map is presented in this work. In a first step, a smooth and dense disparity map is obtained by interpolation of the disparity values obtained from feature-based matching. In a second step, this map is iteratively adjusted, using area-based matching techniques, and a dense disparity map that supports non-smooth surfaces is obtained. The main idea used in this approach is based on biologic evidence of disparity gradient that is present in the living beings vision systems. The methodology was applied to synthetic and real images and the obtained results show the coherence between the 3-D geometry of the scene and the disparity map generated.
%3 slp017.pdf
%O The conference was held in Rio de Janeiro, RJ, Brazil, from October 20 to 23.