Coregistering 3D Models, Range, and Optical Imagery Using Least-Median SquaresFitting

摘要

When using methods for recovering the least squares optimal pose+registrationparameters between a model and an image suite, construction of the feature correspondences is key. Inclusion of outliers in the correspondence set can severly deteriorate the performance and fidelity of these pose+coregistration estimation algorithms. We consider the use of median filtering in the construction of consistent correspondences. Finally, we test our coregistration with a median filtering system on real ATR data.