| Abstract |
As 3D shapes become more common, the need to effectively visualise them
gains importance. This visualisation is typically achieved by presenting a
viewer with a few representative images of the 3D shape. This talk is
divided into three parts, each approaching the above problem from a
different aspect.
Best view methods compare different views of the 3D shape based on
geometric criteria and choose the best one(s) among them. We present a
method that exploits results of perceptive experiments to compute N
representative views of the 3D shape.
View comparison methods typically employ methods that are inherently
rotation invariant. As 3D models do not usually contain up-vector
information, selected views often contain the shape in arbitrary
orientations. We present an example-based method to correct the
orientation of a shape in a view image.
The last part of the talk presents our solution to the "best fly" problem,
which is a natural extension of the best view problem. Instead of finding
static views of the shape, we aim to compute a representative animation of
it. A path for a virtual camera pointing at the shape is computed, and its
speed and zoom along the path are varied according to properties of the
shape in the current view. |
