Section: Scientific Foundations
Geometric calculation and model transformation
Participants : François Sillion, Cyril Soler, Nicolas Holzschuch, Xavier Décoret, Samuel Hornus, Elmar Eisemann, Aurélien Martinet, Lionel Baboud, Thierry Stein.
Geometric models of a 3D scene are available from a variety of sources, including industrial partners. In our experience, more 3D geometry files lack all forms of high level information, either because it was lost during format conversions, or because it was not defined by the designer of the model. On the other hand, most researchers working with 3D scene data would like to use such high level information such as which groups of polygons form connex shapes, human regognisable objects, have symetries, or even which groups of polygons look like each other (also known as instancing information ). We are working on algorithms to automatically retrieve some high level (also named semantic ) information from a polygon soup , i.e a list of polygons without any information about how these polygons are related to each other.
Creating images from three-dimensional models is a computationally –intensive task. A particularly difficult issue has long been the calculation of visibility information in 3D scenes. We are working on several issues related to visibility, such as the decomposition of a scene into appropriate regions (or cells) to assist in the precalculation of visibility relationships, or the precalculation of object sets visible from a particular view point or region of space.
More generally, we are interested in all aspects of geometric calculation that lead to the creation, simplification or transformation of 3D models. Complex scenes for virtual environments are typically assembled using data from very different sources, therefore coming in very different resolutions or amounts of detail. It is often a requirement to suppress unneeded detail in some parts of the scene, or to generate detail where it is missing. Given the very high cost of manual modeling, fully or semi-automated techniques are essential.
Furthermore, the apparent complexity and the amount of detail should also be adapted to the particular usage in the application, and we advocate that this can be realized by choosing appropriate data representations. We are therefore working on innovative data representations for 3D scenes, notably involving many image-based techniques.
