THESIS
2007
viii, 87 leaves : ill. ; 30 cm
Abstract
Three-dimensional (3D) model is still the de facto fundamental component for 3D graphic rendering. While significant advances in techniques in 3D rendering have been made, and extremely realistic and fascinating virtual world can be displayed, creating a detailed 3D model, especially for a physically non-existing object, is still a tedious or very time-consuming task that requires a lot of skills and labors....[
Read more ]
Three-dimensional (3D) model is still the de facto fundamental component for 3D graphic rendering. While significant advances in techniques in 3D rendering have been made, and extremely realistic and fascinating virtual world can be displayed, creating a detailed 3D model, especially for a physically non-existing object, is still a tedious or very time-consuming task that requires a lot of skills and labors.
On the other hand, in computer vision, automatic techniques in stereo reconstruction allow us to capture detail geometric information from the real world. Although stereo vision may not be a direct answer to 3D modeling that satisfies the high quality demand necessary for realistic rendering to reveal subtle geometric details, the scientific merits and mathematics underpinning recent advances in stereo vision inspire useful ideas which may lead to new solution to interactive 3D modeling.
In this thesis, we will present such a new solution that brings together photometric stereo and interactive modeling under the auspices of human-computer interaction. By starting our study on dense photometric stereo, where a large class of objects can be digitized precisely despite the presence of severe shadows, highlight, transparency and complex geometry, high-quality and accurate normal maps can be produced. Such normal map serves as a vehicle to our innovative and effective interactive 3D markup strategy, which we call the ShapePalette metaphor, which constitutes a bridge connecting human and computer so that they can work in tandem to create complex 3D shapes, where the human user is only required to draw a few simple 2D primitives and a detail 3D model can be quickly materialized from line arts, sketches, and photographs. We present impressive results and propose future work on this thesis.
Post a Comment