1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
using namespace cv;
using namespace cv::gpu;
using namespace std;
#if !defined(HAVE_CUDA)
void cv::gpu::transformPoints(const GpuMat&, const Mat&, const Mat&, GpuMat&, Stream&) { throw_nogpu(); }
void cv::gpu::projectPoints(const GpuMat&, const Mat&, const Mat&, const Mat&, const Mat&, GpuMat&, Stream&) { throw_nogpu(); }
void cv::gpu::solvePnPRansac(const Mat&, const Mat&, const Mat&, const Mat&, Mat&, Mat&, bool, int, float, int, vector<int>*) { throw_nogpu(); }
#else
namespace cv { namespace gpu { namespace device
{
namespace transform_points
{
void call(const DevMem2D_<float3> src, const float* rot, const float* transl, DevMem2D_<float3> dst, cudaStream_t stream);
}
namespace project_points
{
void call(const DevMem2D_<float3> src, const float* rot, const float* transl, const float* proj, DevMem2D_<float2> dst, cudaStream_t stream);
}
namespace solve_pnp_ransac
{
int maxNumIters();
void computeHypothesisScores(
const int num_hypotheses, const int num_points, const float* rot_matrices,
const float3* transl_vectors, const float3* object, const float2* image,
const float dist_threshold, int* hypothesis_scores);
}
}}}
using namespace ::cv::gpu::device;
namespace
{
void transformPointsCaller(const GpuMat& src, const Mat& rvec, const Mat& tvec, GpuMat& dst, cudaStream_t stream)
{
CV_Assert(src.rows == 1 && src.cols > 0 && src.type() == CV_32FC3);
CV_Assert(rvec.size() == Size(3, 1) && rvec.type() == CV_32F);
CV_Assert(tvec.size() == Size(3, 1) && tvec.type() == CV_32F);
// Convert rotation vector into matrix
Mat rot;
Rodrigues(rvec, rot);
dst.create(src.size(), src.type());
transform_points::call(src, rot.ptr<float>(), tvec.ptr<float>(), dst, stream);
}
}
void cv::gpu::transformPoints(const GpuMat& src, const Mat& rvec, const Mat& tvec, GpuMat& dst, Stream& stream)
{
transformPointsCaller(src, rvec, tvec, dst, StreamAccessor::getStream(stream));
}
namespace
{
void projectPointsCaller(const GpuMat& src, const Mat& rvec, const Mat& tvec, const Mat& camera_mat, const Mat& dist_coef, GpuMat& dst, cudaStream_t stream)
{
CV_Assert(src.rows == 1 && src.cols > 0 && src.type() == CV_32FC3);
CV_Assert(rvec.size() == Size(3, 1) && rvec.type() == CV_32F);
CV_Assert(tvec.size() == Size(3, 1) && tvec.type() == CV_32F);
CV_Assert(camera_mat.size() == Size(3, 3) && camera_mat.type() == CV_32F);
CV_Assert(dist_coef.empty()); // Undistortion isn't supported
// Convert rotation vector into matrix
Mat rot;
Rodrigues(rvec, rot);
dst.create(src.size(), CV_32FC2);
project_points::call(src, rot.ptr<float>(), tvec.ptr<float>(), camera_mat.ptr<float>(), dst,stream);
}
}
void cv::gpu::projectPoints(const GpuMat& src, const Mat& rvec, const Mat& tvec, const Mat& camera_mat, const Mat& dist_coef, GpuMat& dst, Stream& stream)
{
projectPointsCaller(src, rvec, tvec, camera_mat, dist_coef, dst, StreamAccessor::getStream(stream));
}
namespace
{
// Selects subset_size random different points from [0, num_points - 1] range
void selectRandom(int subset_size, int num_points, vector<int>& subset)
{
subset.resize(subset_size);
for (int i = 0; i < subset_size; ++i)
{
bool was;
do
{
subset[i] = rand() % num_points;
was = false;
for (int j = 0; j < i; ++j)
if (subset[j] == subset[i])
{
was = true;
break;
}
} while (was);
}
}
// Computes rotation, translation pair for small subsets if the input data
class TransformHypothesesGenerator
{
public:
TransformHypothesesGenerator(const Mat& object_, const Mat& image_, const Mat& dist_coef_,
const Mat& camera_mat_, int num_points_, int subset_size_,
Mat rot_matrices_, Mat transl_vectors_)
: object(&object_), image(&image_), dist_coef(&dist_coef_), camera_mat(&camera_mat_),
num_points(num_points_), subset_size(subset_size_), rot_matrices(rot_matrices_),
transl_vectors(transl_vectors_) {}
void operator()(const BlockedRange& range) const
{
// Input data for generation of the current hypothesis
vector<int> subset_indices(subset_size);
Mat_<Point3f> object_subset(1, subset_size);
Mat_<Point2f> image_subset(1, subset_size);
// Current hypothesis data
Mat rot_vec(1, 3, CV_64F);
Mat rot_mat(3, 3, CV_64F);
Mat transl_vec(1, 3, CV_64F);
for (int iter = range.begin(); iter < range.end(); ++iter)
{
selectRandom(subset_size, num_points, subset_indices);
for (int i = 0; i < subset_size; ++i)
{
object_subset(0, i) = object->at<Point3f>(subset_indices[i]);
image_subset(0, i) = image->at<Point2f>(subset_indices[i]);
}
solvePnP(object_subset, image_subset, *camera_mat, *dist_coef, rot_vec, transl_vec);
// Remember translation vector
Mat transl_vec_ = transl_vectors.colRange(iter * 3, (iter + 1) * 3);
transl_vec = transl_vec.reshape(0, 1);
transl_vec.convertTo(transl_vec_, CV_32F);
// Remember rotation matrix
Rodrigues(rot_vec, rot_mat);
Mat rot_mat_ = rot_matrices.colRange(iter * 9, (iter + 1) * 9).reshape(0, 3);
rot_mat.convertTo(rot_mat_, CV_32F);
}
}
const Mat* object;
const Mat* image;
const Mat* dist_coef;
const Mat* camera_mat;
int num_points;
int subset_size;
// Hypotheses storage (global)
Mat rot_matrices;
Mat transl_vectors;
};
}
void cv::gpu::solvePnPRansac(const Mat& object, const Mat& image, const Mat& camera_mat,
const Mat& dist_coef, Mat& rvec, Mat& tvec, bool use_extrinsic_guess,
int num_iters, float max_dist, int min_inlier_count,
vector<int>* inliers)
{
CV_Assert(object.rows == 1 && object.cols > 0 && object.type() == CV_32FC3);
CV_Assert(image.rows == 1 && image.cols > 0 && image.type() == CV_32FC2);
CV_Assert(object.cols == image.cols);
CV_Assert(camera_mat.size() == Size(3, 3) && camera_mat.type() == CV_32F);
CV_Assert(!use_extrinsic_guess); // We don't support initial guess for now
CV_Assert(num_iters <= solve_pnp_ransac::maxNumIters());
const int subset_size = 4;
const int num_points = object.cols;
CV_Assert(num_points >= subset_size);
// Unapply distortion and intrinsic camera transformations
Mat eye_camera_mat = Mat::eye(3, 3, CV_32F);
Mat empty_dist_coef;
Mat image_normalized;
undistortPoints(image, image_normalized, camera_mat, dist_coef, Mat(), eye_camera_mat);
// Hypotheses storage (global)
Mat rot_matrices(1, num_iters * 9, CV_32F);
Mat transl_vectors(1, num_iters * 3, CV_32F);
// Generate set of hypotheses using small subsets of the input data
TransformHypothesesGenerator body(object, image_normalized, empty_dist_coef, eye_camera_mat,
num_points, subset_size, rot_matrices, transl_vectors);
parallel_for(BlockedRange(0, num_iters), body);
// Compute scores (i.e. number of inliers) for each hypothesis
GpuMat d_object(object);
GpuMat d_image_normalized(image_normalized);
GpuMat d_hypothesis_scores(1, num_iters, CV_32S);
solve_pnp_ransac::computeHypothesisScores(
num_iters, num_points, rot_matrices.ptr<float>(), transl_vectors.ptr<float3>(),
d_object.ptr<float3>(), d_image_normalized.ptr<float2>(), max_dist * max_dist,
d_hypothesis_scores.ptr<int>());
// Find the best hypothesis index
Point best_idx;
double best_score;
minMaxLoc(d_hypothesis_scores, NULL, &best_score, NULL, &best_idx);
int num_inliers = static_cast<int>(best_score);
// Extract the best hypothesis data
Mat rot_mat = rot_matrices.colRange(best_idx.x * 9, (best_idx.x + 1) * 9).reshape(0, 3);
Rodrigues(rot_mat, rvec);
rvec = rvec.reshape(0, 1);
tvec = transl_vectors.colRange(best_idx.x * 3, (best_idx.x + 1) * 3).clone();
tvec = tvec.reshape(0, 1);
// Build vector of inlier indices
if (inliers != NULL)
{
inliers->clear();
inliers->reserve(num_inliers);
Point3f p, p_transf;
Point2f p_proj;
const float* rot = rot_mat.ptr<float>();
const float* transl = tvec.ptr<float>();
for (int i = 0; i < num_points; ++i)
{
p = object.at<Point3f>(0, i);
p_transf.x = rot[0] * p.x + rot[1] * p.y + rot[2] * p.z + transl[0];
p_transf.y = rot[3] * p.x + rot[4] * p.y + rot[5] * p.z + transl[1];
p_transf.z = rot[6] * p.x + rot[7] * p.y + rot[8] * p.z + transl[2];
p_proj.x = p_transf.x / p_transf.z;
p_proj.y = p_transf.y / p_transf.z;
if (norm(p_proj - image_normalized.at<Point2f>(0, i)) < max_dist)
inliers->push_back(i);
}
}
}
#endif