Structural matching in 2D using Bayesian methods

This is a statistical method for finding correspondences between two sets of geometric features. Typically one feature set is extracted from an image of some scene, and the other is derived from a model of the same scene.

Starting with an example, we might have an image of the ground taken from an aeroplane or satellite, and the model might be a map of the corresponding road network. There is an unknown Euclidean transformation (i.e. unknown translation and rotation) between image and map:

In order to locate the image on the map, we have to extract some set of features from the image, and also some corresponding features from the map. In this example, the features are line segments corresponding to the roads (shown in green):

The matching algorithm will then find the set of features (shown in magenta) that correspond between image and map. From these correspondences, the unknown Euclidean transformation between image and map can be computed (indicated by the yellow pointers):

To be continued...


Email: Last modified: Thu Jan 27 11:43:31 GMT 2005