SQPnP

C++ Implementation of the SQPnP algorithm.

The algorithm is the generic PnP solver described in the paper "A Consistently Fast and Globally Optimal Solution to the Perspective-n-Point Problem" by G. Terzakis and M. Lourakis. For more intuition, refer to the supplementary material here.

Required libraries

SQPnP requires the Eigen library to build. Besides rank revealing QR and optionally SVD, the use of Eigen is confined to matrix addition, transposition and multiplication. Choosing Eigen was motivated by its increasing popularity and lightweight character. There are also three examples of using the solver in this repository:

1. ) one that uses OpenCV, just for the sake of demonstrating the initialization of SQPnP with cv::Point_<> and cv::Point3_<> structures,
2. ) another in which data points are copied from plain 2D arrays, and
3. ) a third which demonstrates SQPnP within RANSAC.

Build will proceed with either one of 1) or 2), depending on whether OpenCV is found or not. Example 3) uses the RansacLib template-based library, (part of) which is included in this repository.

Build

Create a build directory in the root of the cloned repository and run cmake:

mkdir build

cd build

cmake ..

or, for a release build,

cmake .. -DCMAKE_BUILD_TYPE=Release

The latter will allow for more accurate timing of average execution time. Finally build everything:

make

To run the PnP example, once in the build directory,

./examples/sqpnp_example

To run the robust estimation example, from the build directory, ./examples/robust_sqpnp_example data/K.txt data/32D.txt

Non-default parameters

See struct SolverParameters in types.h which contains SQPnP's parameters that can be specified by the caller. For instance, to use SVD instead of the default RRQR for the nullspace basis of Omega, the following fragment can be used:

  // call solver with user-specified parameters (and equal weights for all points)
  sqpnp::SolverParameters params;
  params.omega_nullspace_method = sqpnp::OmegaNullspaceMethod::SVD;
  sqpnp::PnPSolver solver(points3d, points2d, std::vector<double>(n, 1.0), params);

Similarly, to use SVD in place of FOAM for the nearest rotation matrix computations, use

params.nearest_rotation_method = sqpnp::NearestRotationMethod::SVD;

Cite as

If you use this code in your published work, please cite the following paper:

@inproceedings{terzakis2020SQPnP,
  title={A Consistently Fast and Globally Optimal Solution to the Perspective-n-Point Problem},
  author={George Terzakis and Manolis Lourakis},
  booktitle={European Conference on Computer Vision},
  pages={478--494},
  year={2020},
  publisher={Springer International Publishing}
}

OpenCV

SQPnP has been integrated into OpenCV 4.x as method cv::SOLVEPNP_SQPNP in solvePnP().