Robust Meshes from Multiple Range Maps: Slide 12 of 14.


Click slide for next, or goto previous, first, last slides or back to thumbnail layout.

Click slide for next, or goto previous, or back to thumbnail layout.

We claim that our method is fairly robust, let's see what contributes to it.

Because of the volumetric approach we get a watertight model in the sense that there are no holes in the surface. Holes due to missing data are automatically filled with surface that lies between volumes known to be empty and ones that could be part of the object.

Interval analysis techniques analyze the space conservatively, delaying decisions until there is some conclusive evidence, automatically refining the analysis close to the surface.

Use of directional information allows us to remove outliers. For random outliers that actually lie outside of the object there's likely to be some view that can prove that a cube surrounding the outlier is not part of the object.

We can also recover thin objects.

The main goal was to avoid using non-robust methods such as averaging as much as possible. Instead, we want to propagate and manipulate constraints into more useful forms before trying to accurately fit and simplify the surface representation.