Multi-Texturing for 3D Objects
Combining 3D geometry models with 2D textures from original camera images leads
to hybrid models that might exploit the best features of both worlds.
Sometimes there is only a limited number of original camera images and
the task is than to create a 3D scene model as best as possible.
The following example shows a synthetic 3D geometric model that was textured with a number of original views.
The geometric model was obtained from a number of calibrated input cameras and a hierarchical octree voxel approach.
After transforming the voxel model into a coarse wireframe, some of the original images were mapped onto the model.
First, the model is shown at two original positions. For an animated navigation tour, please click on one of the examples,
although this might load a bit (1,8 MB).
Fig. 1: Two views from original viewpoint
For a preferred natural impression, it is important to provide a form of view-dependent texture mapping.
Although all views are present in the example in Fig. 1, a special interpolation function is applied to each texture
that provides certain visibility aspects: On one hand, the multi-view texture mapping shell only show
a specific texture when the 3D model is observed from the associated original viewpoint.
On the other hand, automatic interpolation has to be applied at intermediate position, preserving the overall illumination intensity.
The latter requirement forces the interpolation algorithm to take into account all original viewing positions within
a certain neighborhood around the current viewpoint. One interpolation example is shown in Fig, 2.
Fig.2: Intermediate view with automatic texture interpolation on HW graphics card
By applying view-dependent multi-texturing, a modeling tool is provided, which preserves illumination conditions
at original viewpoints and provides decent interpolation in between with out the need of extensive artificial light and illumination modeling.
Nevertheless, the quality of interpolated views highly depends on the surface and depth structure of the original model,
the provided original camera images and thus the discrete sampling of the scene.
|
Start
Staff