Pair selection
Candidate image pairs are selected using geographical overlap and stereo baseline constraints.
Abstract
This work presents a generic bundle adjustment methodology for multi-view 3D reconstruction from high-resolution optical satellite imagery. The method refines the Rational Polynomial Camera model of each image using corrections that compensate orientation errors, while feature correspondences provide the tie-point observations used by the joint optimisation.
The pipeline is designed to operate without ground control points and includes image-pair selection, local feature matching, multi-view track construction, RPC correction, and geometric evaluation. The accompanying implementation constitutes the original sat-bundleadjust v1 release.
Method at a glance
Feature correspondences
Handcrafted local features and matching procedures establish pairwise correspondences.
RPC refinement
Multi-view feature tracks constrain a joint bundle adjustment of RPC correction parameters.
Software release
sat-bundleadjust v1.0.0
This release tags the original implementation developed for the 2021 IPOL article.
- RPC model handling based on the rpcm package.
- Feature matching based on handcrafted detectors, descriptors, and matching methods.
- Image-pair selection based on stereo baseline distance and geographical overlap.
- RPC refinement through bundle adjustment using feature tracks as tie points.
- Corrected camera models exported in RPC format using rpcfit.
Citation
@article{mari2021generic,
title = {A Generic Bundle Adjustment Methodology for Indirect
RPC Model Refinement of Satellite Imagery},
author = {Marí, Roger and de Franchis, Carlo and
Meinhardt-Llopis, Enric and Anger, Jérémy and
Facciolo, Gabriele},
journal = {Image Processing On Line},
volume = {11},
pages = {344--373},
year = {2021},
doi = {10.5201/ipol.2021.352}
}