FINAL PROJECT TOPIC:
MANHATTAN PROJECT
PROBLEM:
This continues our interest in the automatic
construction of large scale geometric models
(Manhattan is our chosen target), and the
associated visualization issues.
There are several subproblems to work on:
SUBPROBLEM 1:
Refinement of our original approach.
-- Building are more than just rectangular
boxes. How can we collect and use of textures
in a more general way? E.g., a facade may
be divided into the ground floor texture
and the ``rest''. But we need a systematic
way to specify this for the renderer.
-- Refinement of the texture library
structure, for instance, to support the
more complicated texturing of buildings
in previous item.
-- Automatic generation of street segments
from maps in which each block is a polygon
(not necessary convex). This would allow
us to generate very large realistic neighborhoods
very quickly from available maps.
SUBPROBLEM 2:
Assume that we have acquired such a large model.
How do we construct a visualization system
that allow a user to "walkthrough" or "flyover"
the model in real time?
The basic issue is one of preprocessing the model
to allow rapid display. Note that you may
work with VRML for this.
ISSUES:
-- Visibility Culling
-- Hierarchical LOD model construction
-- How to use foveated images
-- Lighting preprocessing (e.g., use radiosity method)
NOTE: If we have not yet lectured on these topics
please consult us for details or the literature.
TOOLS, RESOURCES:
-- can use the "Large Model" API on SGI machines to
help solve this problem.