Catalogue of Artificial Intelligence Techniques
Aliases: Model Parameter Determination
Categories: Pattern Recognition and Image Processing , Vision
Author(s): David Lowe
An important problem in model-based computer vision systems is to solve for the orientation and position of a three-dimensional object given some matches between features of the object and edges in the image. This algorithm provides a practical solution to the problem, including the difficult case of determining full three-dimensional viewpoint from locations of features in a two-dimensional image. The method is also capable of solving for internal model parameters such as variable sizes and articulations. The solution is based upon providing a rough initial estimate for the unknown parameters and then using Newton iteration to achieve an optimal least-squares fit. An important application is as part of a search process, in which a few matches can be used to determine viewpoint, which in turn constrains the locations of further matches.
- Lowe, D.G., Perceptual Organisation and Visual Recognition
, Kluwer Academic Publishers, Boston, Mass., 1985.