Skip to content

[ICCV 2021] [IJCV] Minimal Solvers for Relative Pose Estimation of Multi-Camera Systems using Affine Correspondences

License

Notifications You must be signed in to change notification settings

jizhaox/relative_pose_gcam_affine

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

10 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

README

This package implements the relative pose estimation for multi-camera systems from affine correspondences: 1AC plane solver, 2AC plane solver and 2AC vertical solver.

Source codes and Matlab mex files with demo code are provided in the package. The core solvers are written by C++. Matlab mex files are compiled using Ubuntu 16.04 + Matlab R2018b. Run test_solver_AC.m in folder "evaluation".

Reference

[1] Banglei Guan, Ji Zhao, Daniel Barath, and Friedrich Fraundorfer. Minimal Solvers for Relative Pose Estimation of Multi-Camera Systems using Affine Correspondences. International Journal of Computer Vision. DOI: https://doi.org/10.1007/s11263-022-01690-w

[1] Banglei Guan, Ji Zhao, Daniel Barath, and Friedrich Fraundorfer. Minimal Cases for Computing the Generalized Relative Pose using Affine Correspondences. International Conference on Computer Vision, 2021.

If you use this package in an academic work, please cite:

@article{guan2023minimal,
  title={Minimal Solvers for Relative Pose Estimation of Multi-Camera Systems using Affine Correspondences},
  author={Guan, Banglei and Zhao, Ji and Barath, Daniel and Fraundorfer, Friedrich},
  journal={International Journal of Computer Vision},
  year={2023},
  doi={https://doi.org/10.1007/s11263-022-01690-w}
}

@inproceedings{guan2021minimal,
  title={Minimal Cases for Computing the Generalized Relative Pose using Affine Correspondences},
  author={Guan, Banglei and Zhao, Ji and Barath, Daniel and Fraundorfer, Friedrich},
  booktitle={International Conference on Computer Vision},
  year={2021},
  pages={6068-6077}
}

1AC plane solver

A numerical solver for the relative pose estimation under planar motion using an affine correspondence. Returns a maximum of 4 solutions.

  • Solver: solver_ac_pose3d_1AC_plane.mexa64

  • API: [Rf1tof2_recover, Tf1tof2_recover] = solver_ac_pose3d_1AC_plane(Image1, Image2, At_33, R_cam, t_cam, Option);

  • Input data for Demo:

    Image1 (3*1 matrix): normalized homogeneous image coordinates of an inter-camera feature point expressed in view 1.

    Image2 (3*1 matrix): normalized homogeneous image coordinates of an inter-camera feature point expressed in view 2.

    At_33 (3*3 matrix): At_33(1:2,1:2) = A, At_33(3,3) = 0, where A is the corresponding 2*2 local affine transformations of the inter-camera feature point.

    R_cam (3*3*2 matrix): extrinsic rotation of two cameras expressed in the reference of the multi-camera system.

    t_cam (3*2 matrix): extrinsic translation of two cameras expressed in the reference of the multi-camera system.

    Option: inter refers inter-camera feature point.

  • Output data for Demo:

    Rf1tof2_recover (3*3*N matrix): real number solutions of rotation matrix, N is the number of real number solutions.

    Tf1tof2_recover (3*N matrix): real number solutions of translation.

2AC plane solver

A numerical solver for the relative pose estimation under planar motion using two affine correspondences. Returns a maximum of 4 solutions.

  • Solver: solver_ac_pose3d_2AC_plane.mexa64

  • API: [Rf1tof2_recover, Tf1tof2_recover] = solver_ac_pose3d_2AC_plane(Image1, Image2, At_33, R_cam, t_cam, Option);

  • Input data for Demo:

    Image1 (3*2 matrix): normalized homogeneous image coordinates of two feature points expressed in view 1.

    Image2 (3*2 matrix): normalized homogeneous image coordinates of two feature points expressed in view 2.

    At_33 (3*3*2 matrix): At_33(1:2,1:2,1) = A1, At_33(3,3,1) = 0, At_33(1:2,1:2,2) = A2, At_33(3,3,2) = 0, where A1 and A2 are the corresponding 2*2 local affine transformations of two feature points.

    R_cam (3*3*2 matrix): extrinsic rotation of two cameras expressed in the reference of the multi-camera system.

    t_cam (3*2 matrix): extrinsic translation of two cameras expressed in the reference of the multi-camera system.

    Option: inter refers inter-camera feature points. intra refers intra-camera feature points.

  • Output data for Demo:

    Rf1tof2_recover (3*3*N matrix): real number solutions of rotation matrix, N is the number of real number solutions.

    Tf1tof2_recover (3*N matrix): real number solutions of translation.

2AC vertical solver

A numerical solver for the relative pose estimation with known vertical direction using two affine correspondences. Returns a maximum of 6 solutions.

  • Solver: solver_ac_pose4d.mexa64

  • API: [Rf1tof2_recover, Tf1tof2_recover] = solver_ac_pose4d( Image1, Image2, At_33, R_cam, t_cam, Option);

  • Input data for Demo:

    Image1 (3*2 matrix): normalized homogeneous image coordinates of two feature points expressed in view 1.

    Image2 (3*2 matrix): normalized homogeneous image coordinates of two feature points expressed in view 2.

    At_33 (3*3*2 matrix): At_33(1:2,1:2,1) = A1, At_33(3,3,1) = 0, At_33(1:2,1:2,2) = A2, At_33(3,3,2) = 0, where A1 and A2 are the corresponding 2*2 local affine transformations of two feature points.

    R_cam (3*3*2 matrix): extrinsic rotation of two cameras expressed in the reference of the multi-camera system.

    t_cam (3*2 matrix): extrinsic translation of two cameras expressed in the reference of the multi-camera system.

    Option: inter refers inter-camera feature points. intra refers intra-camera feature points.

  • Output data for Demo:

    Rf1tof2_recover (3*3*N matrix): real number solutions of rotation matrix, N is the number of real number solutions.

    Tf1tof2_recover (3*N matrix): real number solutions of translation.

Compilation

Dependency: Eigen http://eigen.tuxfamily.org

You can compile the package in Linux like this:

{path}$mkdir build
{path}$cd build
{path}/build$cmake ..
{path}/build$make

Matlab Interface and Evaluation

Compiled files using Ubuntu 16.04 + Matlab R2018b are provided. You can run the package in Matlab.

evaluation/test_solver_AC.m is the demo which shows how to call the APIs in Matlab.

Demo Usage in Folder "test"

test/src/test3d_inter_1AC_plane.cpp , test/src/test3d_inter_2AC_plane.cpp, test/src/test3d_intra_2AC_plane.cpp , test/src/test4d_inter_2AC_vertical.cpp and test/src/test4d_intra_2AC_vertical.cpp are the demos which show how to call the APIs in C++. The input parameters of Demo are stored in folder "test/data".

{path}/build$./test3d_inter_1AC_plane ../data/data3d_inter_1AC_plane.bin
{path}/build$./test3d_inter_2AC_plane ../data/data3d_inter_2AC_plane.bin
{path}/build$./test3d_intra_2AC_plane ../data/data3d_intra_2AC_plane.bin
{path}/build$./test4d_inter_2AC_vertical ../data/data4d_inter_2AC_vertical.bin
{path}/build$./test4d_intra_2AC_vertical ../data/data4d_intra_2AC_vertical.bin

About

[ICCV 2021] [IJCV] Minimal Solvers for Relative Pose Estimation of Multi-Camera Systems using Affine Correspondences

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published