hide
Free keywords:
-
Abstract:
The creation of high quality animations of real-world human actors has long
been a challenging problem in computer graphics. It involves the modeling of
the shape of the virtual actors, creating their motion, and the reproduction of
very fine dynamic
details. In order to render the actor under arbitrary lighting, it is required
that reflectance properties are modeled for each point on the surface. These
steps, that are usually performed manually by professional modelers, are time
consuming and cumbersome.
In this thesis, we show that algorithmic solutions for some of the problems
that arise in the creation of high quality animation of real-world people are
possible using multi-view video data. First, we present a novel spatio-temporal
approach
to create a personalized avatar from multi-view video data of a moving person.
Thereafter, we propose two enhancements to a method that captures human shape,
motion and reflectance properties of amoving human using eightmulti-view video
streams. Afterwards we extend this work, and in order to add very fine dynamic
details to the geometric models, such as wrinkles and folds in the clothing, we
make use of the multi-view video recordings and present a statistical method
that can passively capture the fine-grain details of time-varying scene
geometry. Finally, in order to reconstruct structured shape and animation of
the subject from video, we present a dense 3D correspondence finding method
that enables spatiotemporally coherent reconstruction of surface animations
directly frommulti-view
video data.
These algorithmic solutions can be combined to constitute a complete animation
pipeline for acquisition, reconstruction and rendering of high quality virtual
actors from multi-view video data. They can also be used individually in a
system that require the solution of a specific algorithmic sub-problem. The
results demonstrate that using multi-view video data it is possible to find the
model description that enables realistic appearance of animated virtual actors
under different lighting
conditions and exhibits high quality dynamic details in the geometry.
