| Abstract |
An area of research in the fields of computer vision and computer graphics involves the modelling and rendering of complex 3-D scenes from the real world. These scenes are modelled using 3-D models, that describe the geometry of the environment and are combined with photometric measurements to improve the fidelity of the representation of the real world. User's navigation in such a wide virtual environment requires the succession and morphing of many local geometric models. Their storage and visualization are very demanding as far as the storage capacity and the computational complexity are concerned. Thus, to make user's navigation possible in real time, we need to invoke a model's decimation algorithm to simplify these detailed local models. In the current study, we attempted to reduce the decimation of a 3-D model to a problem of accurate content-adaptive 2-D mesh generation. We use a mesh adaptation algorithm that generates a mesh structure, which is the output of a stereo-matching procedure applied on a pair of images. Furthermore, this image provides a depth estimation of the model's points. The effect of the algorithm consists of selecting a part of the initial image samples and performing Delaunay triangulation to create the 2-D mesh connectivity. Finally, we produce the 3-D coordinates of the selected points and use the algorithm's triangulation output to form the simplified model's surfaces. In addition, we examine general aspects of image representation via mesh modelling, as well as motion estimation and compensation throughout an image sequence by means of a deformable mesh structure. Particularly, we compare different methods of approaching an image function in the interior of a mesh, as well as methods of mesh adaptation to local deformations observed in an image sequence. We also study the use of hierarchical meshes for tracking different kinds of motion in each hierarchical level, reducing the complexity and improving the effectiveness of the tracking procedure.
|
Mesh Adaptation to Stereoscopic Data and Image Sequences
