Intelligent Systems Laboratory
Feature Prediction
Since in most industrial applications
the lighting is either known or can be controlled, we are developing analytical
methods which can predict the lines
which will appear, disappear, or break up, under various view-point and
lighting assumptions. A major aspect of our model feature prediction work
involves the prediction of two-dimensional features from the hierarchical,
relational three-dimensional models. We are developing
the mathematics and computer algorithm for
morphologically predicting line and curve segments independent of the lighting.
This kind of
predictor differs from those used in graphics programs in that it does
not use ray casting which
has to be performed for every pixel in the image and is, therefore, slow.
Instead, we use the
geometry and topology of the three-dimensional object to determine three-
dimensional structures that will give rise to two-dimensional features and to
symbolically instead of iconically, produce the two-dimensional features.
