I'm doing an MSEE thesis on generating a hierarchical multiresolution

terrain data base from Digital Terrain Elevation Data (DTED, regularly

sampled elevation posts). I generate polygons (actually triangles) that

would be used by an image generation system to drape texture over them

and render a scene (maybe for flight simulation). The generation of the

data base of triangles is an offline process. It is hierarchical in

that the root of the tree of triangles is the coarsest representation

(large triangles with relatively large maximum height error). The

next branch of the tree splits parent triangles to create a representation

with a smaller maximum height error. This is continued to reach a desired

accuracy. I concern myself with maintaining a continuous surface (i.e,

no holes), triangles that are not "too slivery" (i.e., long & thin),

and minimum number of triangles required for desired maximum height

error.

The idea is that a scene could be rendered by mosaic-ing triangles

from different depths in the data base depending on the required

accuracy. For terrain at the horizon the coarsest representation

could be used. For terrain closer to the eyepoint, a finer

representation could be used.

I believe that two factors in this process are:

1) Filtering the data to avoid aliasing (I'm using a Wavelet

based filtering process).

2) The acutal triangulation method (currently using a

quaternary triangulation scheme, split a triangle

into four by joining a point along each edge).

My question: Is this (i.e., generating triangles from elevation

posts to be used for scene rendering) considered a solved problem?

Would someone actually use this type of data base?

I'm particularly interested in responses from those users of

high end machines, like SGIs. But please don't reserve

responses/opinions to that perspective.

thanks for your time,

Robert Fuentes

(The University of Texas at Arlington)

Lockheed Martin