Description
Although numerical models may be derived ab initio many CAD modelsare still derived from real-world objects. That is, a real shape is captured
and then converted into a numerical representation using either NURBS or
subdivision. This process is referred to as surface reconstruction.
Approaches to find a subdivision control mesh for a given input shape are
typically based on analysing the given surface and then manually or semi-
automatically fit a control mesh which approximates in the limit the smooth
input shape while respecting differential properties of the surface.
The surface reconstruction we present avoids analysis of the input shape.
Instead we use a genetic algorithm to optimize a set of typical modelling op-
erations to derive a subdivision control mesh which in the limit approximates
the input or target shape. The operations are applied to a simple base object
like a cube or torus and are randomly initialized.
The genetic algorithm tries to optimize the set of modelling operations not
only with respect to the input shape, but also such that a good control mesh
for the target shape is achieved. To achieve this a meaningful fitness functions
has to be defined that takes into account the distance of the subdivided test
shape to the target shape how many operations are applied and how many
vertices the subdivided test shape has in comparison to the target shape. A
well suited test shape, with a high fitness value should minimize all three
measures.
We present our genetic algorithm including metrics for determining the
fitness of a test shape.
| Period | 22 Sept 2022 |
|---|---|
| Event title | 3rd International Conference on Subdivision, Geometric and Algebraic Methods, Isogeometric Analysis and Refinability: SMART 2022 |
| Event type | Conference |
| Location | Rimini, ItalyShow on map |