radialpose

This is a C++ implementation of different minimal solvers for absolute pose estimation with radial distortion. Currently, the following solvers are implemented

• 1 - D(1,0) - 5p -- Larsson et al. ICCV 2019
• 2 - D(2,0) - 5p -- Larsson et al. ICCV 2019
• 3 - D(3,0) - 5p -- Larsson et al. ICCV 2019 (Minimal)
• 4 - D(3,3) - 8p -- Larsson et al. ICCV 2019
• 5 - U(1,0) - 5p -- Larsson et al. ICCV 2019
• 6 - U(0,1) - 4p -- Larsson et al. ICCV 2017 (Minimal, Non-planar)
• 7 - U(0,1) - 4p -- Bujnak et al. ACCV 2010 (Minimal, Non-planar)
• 8 - U(0,1) - 5p -- Kukelova et al. ICCV 2013
• 9 - U(0,2) - 5p -- Kukelova et al. ICCV 2013
• 10 - U(0,3) - 5p -- Kukelova et al. ICCV 2013 (Minimal)
• 11 - U(0,1) - 4p -- Oskarsson arxiv 2018 (Minimal, Planar)
• 12 - N/A - 5p -- Kukelova et al. ICCV 2013 (Minimal, 1D Radial)

where D(2,0) corresponds to a distortion model with two polynomial parameters and zero division parameters, see the paper for more details. Note that the runtime experiments in the paper were done with a more barebones implementation of these solvers.

Installation

The solvers are available as a library. To build run

mkdir build
cd build/
cmake ../
make -j

After building you should be able to run radialpose_test and ransac_test.

There are currently two dependencies:

• Eigen
• RansacLib

There are also Matlab mex interfaces. Check matlab/compile_mex.m.

Using radialpose

The different solvers are available through the classes

• radialpose::larsson_iccv19::Solver<Np,Nd,DistortionModel>()
• radialpose::larsson_iccv17::NonPlanarSolver()
• radialpose::bujnak_accv10::NonPlanarSolver()
• radialpose::kukelova_iccv13::Solver()
• radialpose::kukelova_iccv13::Radial1DSolver(Nd)
• radialpose::oskarsson_arxiv18::PlanarSolver

Each of these solvers implement a function int estimate(const Points2D &image_points, const Points3D &world_points, std::vector<Camera> *poses) const; which should be used to call the solvers. See solvers/pose_estimator.h for definitions of Points2D and Points3D, and see misc/camera.h for the definition of Camera.

Important: You should always call Solver.estimate() instead of Solver.solve(). The estimate function takes care of rescaling the input data before passing it to the solver as well as filtering some bad solutions and other nice things.

RansacLib wrappers

For each solver there is also a RansacLib wrapper in radialpose::RansacEstimator<Solver>.

For example, to do LO-RANSAC with the D(2,0) solver from the ICCV19 paper you can do the following:

larsson_iccv19::Solver<2, 0, true> estimator;
RansacEstimator<larsson_iccv19::Solver<2, 0, true>>  solver(x, X, estimator);

Camera best_model;
ransac_lib::RansacStatistics ransac_stats;

ransac_lib::LocallyOptimizedMSAC<Camera,std::vector<Camera>,RansacEstimator<larsson_iccv19::Solver<2, 0, true>>> lomsac;
int inliers =  lomsac.EstimateModel(options, solver, &best_model, &ransac_stats);

License

radialpose is licensed under the BSD 3-Clause license. Please see License for details.

Citing

If you are using the library for (scientific) publications, please cite the following source:

@inproceedings{larsson2019revisiting,
  title = {{Revisiting Radial Distortion Absolute Pose}},
  author = {Larsson, Viktor and Sattler, Torsten and Kukelova, Zuzana and Pollefeys, Marc},
  booktitle = {International Conference on Computer Vision (ICCV)},
  year = {2019}
}

If you use any of the re-implementations of the other methods please cite the corresponding papers as well.

@inproceedings{bujnak2010new,
  title={New efficient solution to the absolute pose problem for camera with unknown focal length and radial distortion},
  author={Bujnak, Martin and Kukelova, Zuzana and Pajdla, Tomas},
  booktitle={Asian Conference on Computer Vision (ACCV)},
  year={2010}
}
@inproceedings{kukelova2013real,
  title={Real-time solution to the absolute pose problem with unknown radial distortion and focal length},
  author={Kukelova, Zuzana and Bujnak, Martin and Pajdla, Tomas},
  booktitle = {International Conference on Computer Vision (ICCV)},
  year={2013}
}
@inproceedings{larsson2017making,
  title={Making minimal solvers for absolute pose estimation compact and robust},
  author={Larsson, Viktor and Kukelova, Zuzana and Zheng, Yinqiang},
  booktitle = {International Conference on Computer Vision (ICCV)},
  year={2017}
}
@article{oskarsson2018fast,
  title={A fast minimal solver for absolute camera pose with unknown focal length and radial distortion from four planar points},
  author={Oskarsson, Magnus},
  journal={arXiv preprint arXiv:1805.10705},
  year={2018}
}