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: keir@google.com (Keir Mierle)
30 //
31 // A minimal, self-contained bundle adjuster using Ceres, that reads
32 // files from University of Washington' Bundle Adjustment in the Large dataset:
33 // http://grail.cs.washington.edu/projects/bal
34 //
35 // This does not use the best configuration for solving; see the more involved
36 // bundle_adjuster.cc file for details.
37 
38 #include <cmath>
39 #include <cstdio>
40 #include <iostream>
41 
42 #include "ceres/ceres.h"
43 #include "ceres/rotation.h"
44 
45 // Read a Bundle Adjustment in the Large dataset.
46 class BALProblem {
47  public:
~BALProblem()48   ~BALProblem() {
49     delete[] point_index_;
50     delete[] camera_index_;
51     delete[] observations_;
52     delete[] parameters_;
53   }
54 
num_observations() const55   int num_observations()       const { return num_observations_;               }
observations() const56   const double* observations() const { return observations_;                   }
mutable_cameras()57   double* mutable_cameras()          { return parameters_;                     }
mutable_points()58   double* mutable_points()           { return parameters_  + 9 * num_cameras_; }
59 
mutable_camera_for_observation(int i)60   double* mutable_camera_for_observation(int i) {
61     return mutable_cameras() + camera_index_[i] * 9;
62   }
mutable_point_for_observation(int i)63   double* mutable_point_for_observation(int i) {
64     return mutable_points() + point_index_[i] * 3;
65   }
66 
LoadFile(const char * filename)67   bool LoadFile(const char* filename) {
68     FILE* fptr = fopen(filename, "r");
69     if (fptr == NULL) {
70       return false;
71     };
72 
73     FscanfOrDie(fptr, "%d", &num_cameras_);
74     FscanfOrDie(fptr, "%d", &num_points_);
75     FscanfOrDie(fptr, "%d", &num_observations_);
76 
77     point_index_ = new int[num_observations_];
78     camera_index_ = new int[num_observations_];
79     observations_ = new double[2 * num_observations_];
80 
81     num_parameters_ = 9 * num_cameras_ + 3 * num_points_;
82     parameters_ = new double[num_parameters_];
83 
84     for (int i = 0; i < num_observations_; ++i) {
85       FscanfOrDie(fptr, "%d", camera_index_ + i);
86       FscanfOrDie(fptr, "%d", point_index_ + i);
87       for (int j = 0; j < 2; ++j) {
88         FscanfOrDie(fptr, "%lf", observations_ + 2*i + j);
89       }
90     }
91 
92     for (int i = 0; i < num_parameters_; ++i) {
93       FscanfOrDie(fptr, "%lf", parameters_ + i);
94     }
95     return true;
96   }
97 
98  private:
99   template<typename T>
FscanfOrDie(FILE * fptr,const char * format,T * value)100   void FscanfOrDie(FILE *fptr, const char *format, T *value) {
101     int num_scanned = fscanf(fptr, format, value);
102     if (num_scanned != 1) {
103       LOG(FATAL) << "Invalid UW data file.";
104     }
105   }
106 
107   int num_cameras_;
108   int num_points_;
109   int num_observations_;
110   int num_parameters_;
111 
112   int* point_index_;
113   int* camera_index_;
114   double* observations_;
115   double* parameters_;
116 };
117 
118 // Templated pinhole camera model for used with Ceres.  The camera is
119 // parameterized using 9 parameters: 3 for rotation, 3 for translation, 1 for
120 // focal length and 2 for radial distortion. The principal point is not modeled
121 // (i.e. it is assumed be located at the image center).
122 struct SnavelyReprojectionError {
SnavelyReprojectionErrorSnavelyReprojectionError123   SnavelyReprojectionError(double observed_x, double observed_y)
124       : observed_x(observed_x), observed_y(observed_y) {}
125 
126   template <typename T>
operator ()SnavelyReprojectionError127   bool operator()(const T* const camera,
128                   const T* const point,
129                   T* residuals) const {
130     // camera[0,1,2] are the angle-axis rotation.
131     T p[3];
132     ceres::AngleAxisRotatePoint(camera, point, p);
133 
134     // camera[3,4,5] are the translation.
135     p[0] += camera[3];
136     p[1] += camera[4];
137     p[2] += camera[5];
138 
139     // Compute the center of distortion. The sign change comes from
140     // the camera model that Noah Snavely's Bundler assumes, whereby
141     // the camera coordinate system has a negative z axis.
142     T xp = - p[0] / p[2];
143     T yp = - p[1] / p[2];
144 
145     // Apply second and fourth order radial distortion.
146     const T& l1 = camera[7];
147     const T& l2 = camera[8];
148     T r2 = xp*xp + yp*yp;
149     T distortion = T(1.0) + r2  * (l1 + l2  * r2);
150 
151     // Compute final projected point position.
152     const T& focal = camera[6];
153     T predicted_x = focal * distortion * xp;
154     T predicted_y = focal * distortion * yp;
155 
156     // The error is the difference between the predicted and observed position.
157     residuals[0] = predicted_x - T(observed_x);
158     residuals[1] = predicted_y - T(observed_y);
159 
160     return true;
161   }
162 
163   // Factory to hide the construction of the CostFunction object from
164   // the client code.
CreateSnavelyReprojectionError165   static ceres::CostFunction* Create(const double observed_x,
166                                      const double observed_y) {
167     return (new ceres::AutoDiffCostFunction<SnavelyReprojectionError, 2, 9, 3>(
168                 new SnavelyReprojectionError(observed_x, observed_y)));
169   }
170 
171   double observed_x;
172   double observed_y;
173 };
174 
main(int argc,char ** argv)175 int main(int argc, char** argv) {
176   google::InitGoogleLogging(argv[0]);
177   if (argc != 2) {
178     std::cerr << "usage: simple_bundle_adjuster <bal_problem>\n";
179     return 1;
180   }
181 
182   BALProblem bal_problem;
183   if (!bal_problem.LoadFile(argv[1])) {
184     std::cerr << "ERROR: unable to open file " << argv[1] << "\n";
185     return 1;
186   }
187 
188   const double* observations = bal_problem.observations();
189 
190   // Create residuals for each observation in the bundle adjustment problem. The
191   // parameters for cameras and points are added automatically.
192   ceres::Problem problem;
193   for (int i = 0; i < bal_problem.num_observations(); ++i) {
194     // Each Residual block takes a point and a camera as input and outputs a 2
195     // dimensional residual. Internally, the cost function stores the observed
196     // image location and compares the reprojection against the observation.
197 
198     ceres::CostFunction* cost_function =
199         SnavelyReprojectionError::Create(observations[2 * i + 0],
200                                          observations[2 * i + 1]);
201     problem.AddResidualBlock(cost_function,
202                              NULL /* squared loss */,
203                              bal_problem.mutable_camera_for_observation(i),
204                              bal_problem.mutable_point_for_observation(i));
205   }
206 
207   // Make Ceres automatically detect the bundle structure. Note that the
208   // standard solver, SPARSE_NORMAL_CHOLESKY, also works fine but it is slower
209   // for standard bundle adjustment problems.
210   ceres::Solver::Options options;
211   options.linear_solver_type = ceres::DENSE_SCHUR;
212   options.minimizer_progress_to_stdout = true;
213 
214   ceres::Solver::Summary summary;
215   ceres::Solve(options, &problem, &summary);
216   std::cout << summary.FullReport() << "\n";
217   return 0;
218 }
219