Intelligent Systems Laboratory
Matching
We are developing several different strategies for matching
two--dimensional images of three--dimensional objects to three dimensional
hierarchical relational models.
One strategy matches two--dimensional features of the image to
three--dimensional features of the model with the help of a perspective
projection expert. Another strategy employs
two--dimensional models of the major view classes of the object.
Two--dimensional
features from the image are matched to two--dimensional features of the view
classes. Once the appropriate view class is determined, the exact pose of
the object can be determined. A third strategy we are developing is applicable
when viewing relatively flat objects. In this case, an affine transformation
can be used as an approximation for the perspective projection. Matching
can be accomplished using affine invariant features in an algorithm having
cubic complexity in the worst case.
