bundle_adjustment.h

Go to the documentation of this file.

/**
 * Copyright (C) 2010  Hauke Strasdat
 *                     Imperial College London
 * Copyright (c) 2005-2020, Individual contributors, see AUTHORS file
 *   See: https://www.mrpt.org/Authors - All rights reserved.
 *
 * bundle_adjuster.h is part of RobotVision.
 *
 * RobotVision is free software: you can redistribute it and/or modify
 * it under the terms of the GNU Lesser General Public License as
 * published by the Free Software Foundation, either version 3 of the
 * License, or any later version.
 *
 * RobotVision is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
 * GNU Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * and the GNU Lesser General Public License along with this program.
 * If not, see <http://www.gnu.org/licenses/>.
 */

#pragma once

#include <mrpt/img/TCamera.h>
#include <mrpt/math/CVectorDynamic.h>
#include <mrpt/system/TParameters.h>
#include <mrpt/vision/types.h>
#include <array>
#include <functional>

// The methods declared in this file are implemented in separate files in:
// vision/src/ba_*.cpp
namespace mrpt::vision
{
/** \defgroup bundle_adj Bundle-Adjustment methods
 * \ingroup mrpt_vision_grp
 */

/** @name Bundle-Adjustment methods
    @{ */

/** A functor type for BA methods \sa bundle_adj_full */
using TBundleAdjustmentFeedbackFunctor = std::function<void(
    const size_t cur_iter, const double cur_total_sq_error,
    const size_t max_iters,
    const mrpt::vision::TSequenceFeatureObservations& input_observations,
    const mrpt::vision::TFramePosesVec& current_frame_estimate,
    const mrpt::vision::TLandmarkLocationsVec& current_landmark_estimate)>;

/** Sparse Levenberg-Marquart solution to bundle adjustment - optimizes all the
 *camera frames & the landmark locations.
 * At input a gross estimation of the frame poses & the landmark points must be
 *supplied. If you don't have such a
 *  starting point, use mrpt::vision::ba_initial_estimate() to compute it.
 *
 * At output the best found solution will be returned in the variables.
 *Optionally, a functor can be passed for having
 *  feedback on the progress at each iteration (you can use it to refresh a
 *GUI, display a progress bar, etc...).
 *
 * This implementation is almost entirely an adapted version from RobotVision
 *(at OpenSLAM.org) (C) by Hauke Strasdat (Imperial College London), licensed
 *under GNU LGPL.
 *  See the related paper:  H. Strasdat, J.M.M. Montiel, A.J. Davison: "Scale
 *Drift-Aware Large Scale Monocular SLAM", RSS2010,
 *http://www.roboticsproceedings.org/rss06/p10.html
 *
 *  List of optional parameters in "extra_params":
 *      - "verbose" : Verbose output (default=0)
 *      - "max_iterations": Maximum number of iterations to run (default=50)
 *      - "robust_kernel": If !=0, use a robust kernel against outliers
 *(default=1)
 *      - "kernel_param": The pseudo-huber kernel parameter (default=3)
 *      - "mu": Initial mu for LevMarq (default=-1 -> autoguess)
 *      - "num_fix_frames": Number of first frame poses to don't optimize (keep
 *unmodified as they come in)  (default=1: the first pose is the reference and
 *is not modified)
 *      - "num_fix_points": Idem, for the landmarks positions (default=0:
 *optimize
 *all)
 *      - "profiler": If !=0, displays profiling information to the console at
 *return.
 *
 * \note In this function, all coordinates are absolute. Camera frames are such
 *that +Z points forward from the focal point (see the figure in
 *mrpt::obs::CObservationImage).
 * \note The first frame pose will be not updated since at least one frame must
 *remain fixed.
 *
 * \param observations [IN] All the feature observations (WITHOUT distortion),
 *indexed by feature ID as lists of <frame_ID, (x,y)>. See
 *TSequenceFeatureObservations.
 * \param camera_params [IN] The camera parameters, mainly used for the
 *intrinsic 3x3 matrix. Distortion params are ignored since it's assumed that
 *\a observations are already undistorted pixels.
 * \param frame_poses [IN/OUT] Input: Gross estimation of each frame poses.
 *Output: The found optimal solution.
 * \param landmark_points [IN/OUT] Input: Gross estimation of each landmark
 *point. Output: The found optimal solution.
 * \param extra_params [IN] Optional extra parameters. Read above.
 * \param user_feedback [IN] If provided, this functor will be called at each
 *iteration to provide a feedback to the user.
 *
 * \return The final overall squared error.
 * \ingroup bundle_adj
 */
double bundle_adj_full(
    const mrpt::vision::TSequenceFeatureObservations& observations,
    const mrpt::img::TCamera& camera_params,
    mrpt::vision::TFramePosesVec& frame_poses,
    mrpt::vision::TLandmarkLocationsVec& landmark_points,
    const mrpt::system::TParametersDouble& extra_params =
        mrpt::system::TParametersDouble(),
    const mrpt::vision::TBundleAdjustmentFeedbackFunctor user_feedback =
        mrpt::vision::TBundleAdjustmentFeedbackFunctor());

/** @} */

/** @name Bundle-Adjustment Auxiliary methods
    @{ */

/** Fills the frames & landmark points maps with an initial gross estimate from
 * the sequence \a observations, so they can be fed to bundle adjustment
 * methods.
 * \sa bundle_adj_full
 * \ingroup bundle_adj
 */
void ba_initial_estimate(
    const mrpt::vision::TSequenceFeatureObservations& observations,
    const mrpt::img::TCamera& camera_params,
    mrpt::vision::TFramePosesVec& frame_poses,
    mrpt::vision::TLandmarkLocationsVec& landmark_points);

//! \overload
void ba_initial_estimate(
    const mrpt::vision::TSequenceFeatureObservations& observations,
    const mrpt::img::TCamera& camera_params,
    mrpt::vision::TFramePosesMap& frame_poses,
    mrpt::vision::TLandmarkLocationsMap& landmark_points);

/** Compute reprojection error vector (used from within Bundle Adjustment
 * methods, but can be used in general)
 *  See mrpt::vision::bundle_adj_full for a description of most parameters.
 * \param frame_poses_are_inverse If set to true, global camera poses are \f$
 * \ominus F \f$ instead of \f$ F \f$, for each F in frame_poses.
 *
 *  \return Overall squared reprojection error.
 * \ingroup bundle_adj
 */
double reprojectionResiduals(
    const mrpt::vision::TSequenceFeatureObservations& observations,
    const mrpt::img::TCamera& camera_params,
    const mrpt::vision::TFramePosesVec& frame_poses,
    const mrpt::vision::TLandmarkLocationsVec& landmark_points,
    std::vector<std::array<double, 2>>& out_residuals,
    const bool frame_poses_are_inverse, const bool use_robust_kernel = true,
    const double kernel_param = 3.0,
    std::vector<double>* out_kernel_1st_deriv = nullptr);

//! \overload
double reprojectionResiduals(
    const mrpt::vision::TSequenceFeatureObservations& observations,
    const mrpt::img::TCamera& camera_params,
    const mrpt::vision::TFramePosesMap& frame_poses,
    const mrpt::vision::TLandmarkLocationsMap& landmark_points,
    std::vector<std::array<double, 2>>& out_residuals,
    const bool frame_poses_are_inverse, const bool use_robust_kernel = true,
    const double kernel_param = 3.0,
    std::vector<double>* out_kernel_1st_deriv = nullptr);

/** For each pose in the vector \a frame_poses, adds a "delta" increment to the
 * manifold, with the "delta" given in the se(3) Lie algebra:
 *
 *    new_frame_poses[i] = frame_poses[i] [+]
 * delta[(first_idx+6*i):(first_idx+6*i+5)]    , for every pose in \a
 * frame_poses
 *
 *  With the left-multiplication convention of the manifold exp(delta)
 * operator, that is:
 *
 *     p <-- p [+] delta ==>  p <-- exp(delta) * p
 *
 * \param delta_num_vals Used just for sanity check, must be equal to
 * "frame_poses.size() * 6"
 * \ingroup bundle_adj
 */
void add_se3_deltas_to_frames(
    const mrpt::vision::TFramePosesVec& frame_poses,
    const mrpt::math::CVectorDouble& delta, const size_t delta_first_idx,
    const size_t delta_num_vals, mrpt::vision::TFramePosesVec& new_frame_poses,
    const size_t num_fix_frames);

/** For each pose in the vector \a frame_poses, adds a "delta" increment to the
 * manifold, with the "delta" given in the se(3) Lie algebra:
 *
 *    new_landmark_points[i] = landmark_points[i] +
 * delta[(first_idx+3*i):(first_idx+3*i+2)]    , for every pose in \a
 * landmark_points
 *
 * \param delta_num_vals Used just for sanity check, must be equal to
 * "landmark_points.size() * 3"
 * \ingroup bundle_adj
 */
void add_3d_deltas_to_points(
    const mrpt::vision::TLandmarkLocationsVec& landmark_points,
    const mrpt::math::CVectorDouble& delta, const size_t delta_first_idx,
    const size_t delta_num_vals,
    mrpt::vision::TLandmarkLocationsVec& new_landmark_points,
    const size_t num_fix_points);

/** @} */
}  // namespace mrpt::vision