Sat-BundleAdjust

RPC Bundle Adjustment for Multi-View and Multi-Date Satellite Imagery

Sat-BundleAdjust is an open-source Python package for RPC bundle adjustment of satellite imagery. It provides tools for refining RPC camera models from multi-view feature correspondences and supports experiments with multi-date satellite image collections.

Project website: https://centreborelli.github.io/sat-bundleadjust/

Releases:

• SAT-BA v1 (2021): A Generic Bundle Adjustment Methodology for Indirect RPC Model Refinement of Satellite Imagery. Roger Marí, Carlo de Franchis, Enric Meinhardt-Llopis, Jeremy Anger, Gabriele Facciolo. IPOL. [Paper] [Code]
• SAT-BA v2 (2026): Robust RPC Bundle Adjustment for Multi-Date Satellite Imagery with Season-Invariant Correspondences. Roger Marí, Elías Masquil, Xavier Bou, Thibaud Ehret, Gabriele Facciolo. ECCV Workshops. [Paper] [Code]

Installation

Install the bundle_adjust package:

git clone https://github.com/centreborelli/sat-bundleadjust.git
cd sat-bundleadjust
pip install -e .

Check that the installation was successful by running

pytest tests/test_ba.py

Usage

To run the code:

bundle_adjust config.json

where config.json contains a Python dictionary specifying the paths to the input data and any additional configuration parameters.

Default configuration

To run the default configuration use a config.json as follows:

{
  "geotiff_dir": "your/path/to/the/input/geotiff/images",
  "rpc_dir": "your/path/to/the/input/RPC/models",
  "rpc_src": "txt",
  "output_dir": "your/output/path"
}

where:

• geotiff_dir points to the directory containing all the input geotiff image files, with extension .tif
• rpc_dir points to the directory containing all the input RPC camera models, in txt files with extension .rpc. The rpcm package is used to represent RPC models, which can be written to txt files using rpcm.RPCModel.write_to_file.
• rpc_src is a string that can be either "txt", "json" or "geotiff". If "geotiff" is used, then the input RPC models are directly read from the input geotiff image files.
• The output RPC models are written in a folder named rpcs_adj, which is created in the output_dir.

Customized configuration

It is possible to change the BA configuration by adding other parameters to config.json. Listed below are the more important ones:

Global configuration parameters

The majority of these parameters are also commented in ba_pipeline.BundleAdjustmentPipeline

Parameter	Type	Allowed values	Default	Description
ba_method	string	"ba_bruteforce", "ba_global", "ba_sequential"	"ba_brutefoce"	BA strategy. By using "ba_bruteforce" all cameras are adjusted at once without considering the acquisition date of the images. The other values are experimental, for more info check ba_timeseries.run_BA_for_RPC_refinement
cam_model	string	"rpc", "affine", "perspective"	"rpc"	Camera model used at internal level to run the BA. Attention, this is different from the output camera model format, which always follows the RPC standard.
aoi	dict	"rpc", "affine", "perspective"	None	Area of interest where RPC have to be consistent in GeoJSON format and longitude and latitude coordinates. If None, it is computed from the union of all image footprints.
correction_params	list of strings	R, T, K, COMMON_K	["R"]	Correction parameters: R (rotation), T (translation), K (calibration matrix) or COMMON_K to fix K in all cams if K is also in the list. You combine these but only increasingly. For instance, T cannot be used alone, you should use ["R", "T"].
clean_outliers	boolean	True, False	True	Set to True to filter potential outlier feature observations after some initial iterations.
save_figures	boolean	True, False	True	Set to True to save illustration png images of the reprojection errors before and after.

Feature track configuration parameters

Check feature_tracks/ft_utils.init_feature_tracks_config for the list of parameters that can be added to config.json to customize the feature tracking stage of the pipeline.

Check ba_pipeline.__init__ for the list of parameters that can be added to config.json to customize the bundle adjustment pipeline.

Test data

Examples:

bundle_adjust tests/config1.json

Comparison with other methods

The companion repository rogermm14/eval_sat-bundleadjust provides scripts and notebooks for comparing RPC bundle adjustment pipelines on the DFC2019 WorldView-3 multi-date imagery over the Omaha and Jacksonville areas of interest.

The evaluation repository supports comparisons between RPC camera models output by:

• AMES Stereo Pipeline
• SAT-BA v1
• SAT-BA v2