1// Ceres Solver - A fast non-linear least squares minimizer 2// Copyright 2010, 2011, 2012 Google Inc. All rights reserved. 3// http://code.google.com/p/ceres-solver/ 4// 5// Redistribution and use in source and binary forms, with or without 6// modification, are permitted provided that the following conditions are met: 7// 8// * Redistributions of source code must retain the above copyright notice, 9// this list of conditions and the following disclaimer. 10// * Redistributions in binary form must reproduce the above copyright notice, 11// this list of conditions and the following disclaimer in the documentation 12// and/or other materials provided with the distribution. 13// * Neither the name of Google Inc. nor the names of its contributors may be 14// used to endorse or promote products derived from this software without 15// specific prior written permission. 16// 17// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" 18// AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19// IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20// ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE 21// LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 22// CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 23// SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 24// INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 25// CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 26// ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 27// POSSIBILITY OF SUCH DAMAGE. 28// 29// Author: sameeragarwal@google.com (Sameer Agarwal) 30// 31// Templated struct implementing the camera model and residual 32// computation for bundle adjustment used by Noah Snavely's Bundler 33// SfM system. This is also the camera model/residual for the bundle 34// adjustment problems in the BAL dataset. It is templated so that we 35// can use Ceres's automatic differentiation to compute analytic 36// jacobians. 37// 38// For details see: http://phototour.cs.washington.edu/bundler/ 39// and http://grail.cs.washington.edu/projects/bal/ 40 41#ifndef CERES_EXAMPLES_SNAVELY_REPROJECTION_ERROR_H_ 42#define CERES_EXAMPLES_SNAVELY_REPROJECTION_ERROR_H_ 43 44#include "ceres/rotation.h" 45 46namespace ceres { 47namespace examples { 48 49// Templated pinhole camera model for used with Ceres. The camera is 50// parameterized using 9 parameters: 3 for rotation, 3 for translation, 1 for 51// focal length and 2 for radial distortion. The principal point is not modeled 52// (i.e. it is assumed be located at the image center). 53struct SnavelyReprojectionError { 54 SnavelyReprojectionError(double observed_x, double observed_y) 55 : observed_x(observed_x), observed_y(observed_y) {} 56 57 template <typename T> 58 bool operator()(const T* const camera, 59 const T* const point, 60 T* residuals) const { 61 // camera[0,1,2] are the angle-axis rotation. 62 T p[3]; 63 ceres::AngleAxisRotatePoint(camera, point, p); 64 65 // camera[3,4,5] are the translation. 66 p[0] += camera[3]; 67 p[1] += camera[4]; 68 p[2] += camera[5]; 69 70 // Compute the center of distortion. The sign change comes from 71 // the camera model that Noah Snavely's Bundler assumes, whereby 72 // the camera coordinate system has a negative z axis. 73 const T& focal = camera[6]; 74 T xp = - p[0] / p[2]; 75 T yp = - p[1] / p[2]; 76 77 // Apply second and fourth order radial distortion. 78 const T& l1 = camera[7]; 79 const T& l2 = camera[8]; 80 T r2 = xp*xp + yp*yp; 81 T distortion = T(1.0) + r2 * (l1 + l2 * r2); 82 83 // Compute final projected point position. 84 T predicted_x = focal * distortion * xp; 85 T predicted_y = focal * distortion * yp; 86 87 // The error is the difference between the predicted and observed position. 88 residuals[0] = predicted_x - T(observed_x); 89 residuals[1] = predicted_y - T(observed_y); 90 91 return true; 92 } 93 94 // Factory to hide the construction of the CostFunction object from 95 // the client code. 96 static ceres::CostFunction* Create(const double observed_x, 97 const double observed_y) { 98 return (new ceres::AutoDiffCostFunction<SnavelyReprojectionError, 2, 9, 3>( 99 new SnavelyReprojectionError(observed_x, observed_y))); 100 } 101 102 double observed_x; 103 double observed_y; 104}; 105 106// Templated pinhole camera model for used with Ceres. The camera is 107// parameterized using 10 parameters. 4 for rotation, 3 for 108// translation, 1 for focal length and 2 for radial distortion. The 109// principal point is not modeled (i.e. it is assumed be located at 110// the image center). 111struct SnavelyReprojectionErrorWithQuaternions { 112 // (u, v): the position of the observation with respect to the image 113 // center point. 114 SnavelyReprojectionErrorWithQuaternions(double observed_x, double observed_y) 115 : observed_x(observed_x), observed_y(observed_y) {} 116 117 template <typename T> 118 bool operator()(const T* const camera_rotation, 119 const T* const camera_translation_and_intrinsics, 120 const T* const point, 121 T* residuals) const { 122 const T& focal = camera_translation_and_intrinsics[3]; 123 const T& l1 = camera_translation_and_intrinsics[4]; 124 const T& l2 = camera_translation_and_intrinsics[5]; 125 126 // Use a quaternion rotation that doesn't assume the quaternion is 127 // normalized, since one of the ways to run the bundler is to let Ceres 128 // optimize all 4 quaternion parameters unconstrained. 129 T p[3]; 130 QuaternionRotatePoint(camera_rotation, point, p); 131 132 p[0] += camera_translation_and_intrinsics[0]; 133 p[1] += camera_translation_and_intrinsics[1]; 134 p[2] += camera_translation_and_intrinsics[2]; 135 136 // Compute the center of distortion. The sign change comes from 137 // the camera model that Noah Snavely's Bundler assumes, whereby 138 // the camera coordinate system has a negative z axis. 139 T xp = - p[0] / p[2]; 140 T yp = - p[1] / p[2]; 141 142 // Apply second and fourth order radial distortion. 143 T r2 = xp*xp + yp*yp; 144 T distortion = T(1.0) + r2 * (l1 + l2 * r2); 145 146 // Compute final projected point position. 147 T predicted_x = focal * distortion * xp; 148 T predicted_y = focal * distortion * yp; 149 150 // The error is the difference between the predicted and observed position. 151 residuals[0] = predicted_x - T(observed_x); 152 residuals[1] = predicted_y - T(observed_y); 153 154 return true; 155 } 156 157 // Factory to hide the construction of the CostFunction object from 158 // the client code. 159 static ceres::CostFunction* Create(const double observed_x, 160 const double observed_y) { 161 return (new ceres::AutoDiffCostFunction< 162 SnavelyReprojectionErrorWithQuaternions, 2, 4, 6, 3>( 163 new SnavelyReprojectionErrorWithQuaternions(observed_x, 164 observed_y))); 165 } 166 167 double observed_x; 168 double observed_y; 169}; 170 171} // namespace examples 172} // namespace ceres 173 174#endif // CERES_EXAMPLES_SNAVELY_REPROJECTION_ERROR_H_ 175