1/*M///////////////////////////////////////////////////////////////////////////////////////
2//
3//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
4//
5//  By downloading, copying, installing or using the software you agree to this license.
6//  If you do not agree to this license, do not download, install,
7//  copy or use the software.
8//
9//
10//                           License Agreement
11//                For Open Source Computer Vision Library
12//
13// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
14// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
15// Third party copyrights are property of their respective owners.
16//
17// Redistribution and use in source and binary forms, with or without modification,
18// are permitted provided that the following conditions are met:
19//
20//   * Redistribution's of source code must retain the above copyright notice,
21//     this list of conditions and the following disclaimer.
22//
23//   * Redistribution's in binary form must reproduce the above copyright notice,
24//     this list of conditions and the following disclaimer in the documentation
25//     and/or other materials provided with the distribution.
26//
27//   * The name of the copyright holders may not be used to endorse or promote products
28//     derived from this software without specific prior written permission.
29//
30// This software is provided by the copyright holders and contributors "as is" and
31// any express or implied warranties, including, but not limited to, the implied
32// warranties of merchantability and fitness for a particular purpose are disclaimed.
33// In no event shall the Intel Corporation or contributors be liable for any direct,
34// indirect, incidental, special, exemplary, or consequential damages
35// (including, but not limited to, procurement of substitute goods or services;
36// loss of use, data, or profits; or business interruption) however caused
37// and on any theory of liability, whether in contract, strict liability,
38// or tort (including negligence or otherwise) arising in any way out of
39// the use of this software, even if advised of the possibility of such damage.
40//
41//M*/
42
43#include "precomp.hpp"
44
45using namespace cv;
46using namespace cv::cuda;
47
48#if !defined HAVE_CUDA || !defined HAVE_OPENCV_CALIB3D || defined(CUDA_DISABLER)
49
50void cv::cuda::transformPoints(const GpuMat&, const Mat&, const Mat&, GpuMat&, Stream&) { throw_no_cuda(); }
51
52void cv::cuda::projectPoints(const GpuMat&, const Mat&, const Mat&, const Mat&, const Mat&, GpuMat&, Stream&) { throw_no_cuda(); }
53
54void cv::cuda::solvePnPRansac(const Mat&, const Mat&, const Mat&, const Mat&, Mat&, Mat&, bool, int, float, int, std::vector<int>*) { throw_no_cuda(); }
55
56#else
57
58namespace cv { namespace cuda { namespace device
59{
60    namespace transform_points
61    {
62        void call(const PtrStepSz<float3> src, const float* rot, const float* transl, PtrStepSz<float3> dst, cudaStream_t stream);
63    }
64
65    namespace project_points
66    {
67        void call(const PtrStepSz<float3> src, const float* rot, const float* transl, const float* proj, PtrStepSz<float2> dst, cudaStream_t stream);
68    }
69
70    namespace solve_pnp_ransac
71    {
72        int maxNumIters();
73
74        void computeHypothesisScores(
75                const int num_hypotheses, const int num_points, const float* rot_matrices,
76                const float3* transl_vectors, const float3* object, const float2* image,
77                const float dist_threshold, int* hypothesis_scores);
78    }
79}}}
80
81using namespace ::cv::cuda::device;
82
83namespace
84{
85    void transformPointsCaller(const GpuMat& src, const Mat& rvec, const Mat& tvec, GpuMat& dst, cudaStream_t stream)
86    {
87        CV_Assert(src.rows == 1 && src.cols > 0 && src.type() == CV_32FC3);
88        CV_Assert(rvec.size() == Size(3, 1) && rvec.type() == CV_32F);
89        CV_Assert(tvec.size() == Size(3, 1) && tvec.type() == CV_32F);
90
91        // Convert rotation vector into matrix
92        Mat rot;
93        Rodrigues(rvec, rot);
94
95        dst.create(src.size(), src.type());
96        transform_points::call(src, rot.ptr<float>(), tvec.ptr<float>(), dst, stream);
97    }
98}
99
100void cv::cuda::transformPoints(const GpuMat& src, const Mat& rvec, const Mat& tvec, GpuMat& dst, Stream& stream)
101{
102    transformPointsCaller(src, rvec, tvec, dst, StreamAccessor::getStream(stream));
103}
104
105namespace
106{
107    void projectPointsCaller(const GpuMat& src, const Mat& rvec, const Mat& tvec, const Mat& camera_mat, const Mat& dist_coef, GpuMat& dst, cudaStream_t stream)
108    {
109        CV_Assert(src.rows == 1 && src.cols > 0 && src.type() == CV_32FC3);
110        CV_Assert(rvec.size() == Size(3, 1) && rvec.type() == CV_32F);
111        CV_Assert(tvec.size() == Size(3, 1) && tvec.type() == CV_32F);
112        CV_Assert(camera_mat.size() == Size(3, 3) && camera_mat.type() == CV_32F);
113        CV_Assert(dist_coef.empty()); // Undistortion isn't supported
114
115        // Convert rotation vector into matrix
116        Mat rot;
117        Rodrigues(rvec, rot);
118
119        dst.create(src.size(), CV_32FC2);
120        project_points::call(src, rot.ptr<float>(), tvec.ptr<float>(), camera_mat.ptr<float>(), dst,stream);
121    }
122}
123
124void cv::cuda::projectPoints(const GpuMat& src, const Mat& rvec, const Mat& tvec, const Mat& camera_mat, const Mat& dist_coef, GpuMat& dst, Stream& stream)
125{
126    projectPointsCaller(src, rvec, tvec, camera_mat, dist_coef, dst, StreamAccessor::getStream(stream));
127}
128
129namespace
130{
131    // Selects subset_size random different points from [0, num_points - 1] range
132    void selectRandom(int subset_size, int num_points, std::vector<int>& subset)
133    {
134        subset.resize(subset_size);
135        for (int i = 0; i < subset_size; ++i)
136        {
137            bool was;
138            do
139            {
140                subset[i] = rand() % num_points;
141                was = false;
142                for (int j = 0; j < i; ++j)
143                    if (subset[j] == subset[i])
144                    {
145                        was = true;
146                        break;
147                    }
148            } while (was);
149        }
150    }
151
152    // Computes rotation, translation pair for small subsets if the input data
153    class TransformHypothesesGenerator : public ParallelLoopBody
154    {
155    public:
156        TransformHypothesesGenerator(const Mat& object_, const Mat& image_, const Mat& dist_coef_,
157                                     const Mat& camera_mat_, int num_points_, int subset_size_,
158                                     Mat rot_matrices_, Mat transl_vectors_)
159                : object(&object_), image(&image_), dist_coef(&dist_coef_), camera_mat(&camera_mat_),
160                  num_points(num_points_), subset_size(subset_size_), rot_matrices(rot_matrices_),
161                  transl_vectors(transl_vectors_) {}
162
163        void operator()(const Range& range) const
164        {
165            // Input data for generation of the current hypothesis
166            std::vector<int> subset_indices(subset_size);
167            Mat_<Point3f> object_subset(1, subset_size);
168            Mat_<Point2f> image_subset(1, subset_size);
169
170            // Current hypothesis data
171            Mat rot_vec(1, 3, CV_64F);
172            Mat rot_mat(3, 3, CV_64F);
173            Mat transl_vec(1, 3, CV_64F);
174
175            for (int iter = range.start; iter < range.end; ++iter)
176            {
177                selectRandom(subset_size, num_points, subset_indices);
178                for (int i = 0; i < subset_size; ++i)
179                {
180                   object_subset(0, i) = object->at<Point3f>(subset_indices[i]);
181                   image_subset(0, i) = image->at<Point2f>(subset_indices[i]);
182                }
183
184                solvePnP(object_subset, image_subset, *camera_mat, *dist_coef, rot_vec, transl_vec);
185
186                // Remember translation vector
187                Mat transl_vec_ = transl_vectors.colRange(iter * 3, (iter + 1) * 3);
188                transl_vec = transl_vec.reshape(0, 1);
189                transl_vec.convertTo(transl_vec_, CV_32F);
190
191                // Remember rotation matrix
192                Rodrigues(rot_vec, rot_mat);
193                Mat rot_mat_ = rot_matrices.colRange(iter * 9, (iter + 1) * 9).reshape(0, 3);
194                rot_mat.convertTo(rot_mat_, CV_32F);
195            }
196        }
197
198        const Mat* object;
199        const Mat* image;
200        const Mat* dist_coef;
201        const Mat* camera_mat;
202        int num_points;
203        int subset_size;
204
205        // Hypotheses storage (global)
206        Mat rot_matrices;
207        Mat transl_vectors;
208    };
209}
210
211void cv::cuda::solvePnPRansac(const Mat& object, const Mat& image, const Mat& camera_mat,
212                             const Mat& dist_coef, Mat& rvec, Mat& tvec, bool use_extrinsic_guess,
213                             int num_iters, float max_dist, int min_inlier_count,
214                             std::vector<int>* inliers)
215{
216    (void)min_inlier_count;
217    CV_Assert(object.rows == 1 && object.cols > 0 && object.type() == CV_32FC3);
218    CV_Assert(image.rows == 1 && image.cols > 0 && image.type() == CV_32FC2);
219    CV_Assert(object.cols == image.cols);
220    CV_Assert(camera_mat.size() == Size(3, 3) && camera_mat.type() == CV_32F);
221    CV_Assert(!use_extrinsic_guess); // We don't support initial guess for now
222    CV_Assert(num_iters <= solve_pnp_ransac::maxNumIters());
223
224    const int subset_size = 4;
225    const int num_points = object.cols;
226    CV_Assert(num_points >= subset_size);
227
228    // Unapply distortion and intrinsic camera transformations
229    Mat eye_camera_mat = Mat::eye(3, 3, CV_32F);
230    Mat empty_dist_coef;
231    Mat image_normalized;
232    undistortPoints(image, image_normalized, camera_mat, dist_coef, Mat(), eye_camera_mat);
233
234    // Hypotheses storage (global)
235    Mat rot_matrices(1, num_iters * 9, CV_32F);
236    Mat transl_vectors(1, num_iters * 3, CV_32F);
237
238    // Generate set of hypotheses using small subsets of the input data
239    TransformHypothesesGenerator body(object, image_normalized, empty_dist_coef, eye_camera_mat,
240                                      num_points, subset_size, rot_matrices, transl_vectors);
241    parallel_for_(Range(0, num_iters), body);
242
243    // Compute scores (i.e. number of inliers) for each hypothesis
244    GpuMat d_object(object);
245    GpuMat d_image_normalized(image_normalized);
246    GpuMat d_hypothesis_scores(1, num_iters, CV_32S);
247    solve_pnp_ransac::computeHypothesisScores(
248            num_iters, num_points, rot_matrices.ptr<float>(), transl_vectors.ptr<float3>(),
249            d_object.ptr<float3>(), d_image_normalized.ptr<float2>(), max_dist * max_dist,
250            d_hypothesis_scores.ptr<int>());
251
252    // Find the best hypothesis index
253    Point best_idx;
254    double best_score;
255    cuda::minMaxLoc(d_hypothesis_scores, NULL, &best_score, NULL, &best_idx);
256    int num_inliers = static_cast<int>(best_score);
257
258    // Extract the best hypothesis data
259
260    Mat rot_mat = rot_matrices.colRange(best_idx.x * 9, (best_idx.x + 1) * 9).reshape(0, 3);
261    Rodrigues(rot_mat, rvec);
262    rvec = rvec.reshape(0, 1);
263
264    tvec = transl_vectors.colRange(best_idx.x * 3, (best_idx.x + 1) * 3).clone();
265    tvec = tvec.reshape(0, 1);
266
267    // Build vector of inlier indices
268    if (inliers != NULL)
269    {
270        inliers->clear();
271        inliers->reserve(num_inliers);
272
273        Point3f p, p_transf;
274        Point2f p_proj;
275        const float* rot = rot_mat.ptr<float>();
276        const float* transl = tvec.ptr<float>();
277
278        for (int i = 0; i < num_points; ++i)
279        {
280            p = object.at<Point3f>(0, i);
281            p_transf.x = rot[0] * p.x + rot[1] * p.y + rot[2] * p.z + transl[0];
282            p_transf.y = rot[3] * p.x + rot[4] * p.y + rot[5] * p.z + transl[1];
283            p_transf.z = rot[6] * p.x + rot[7] * p.y + rot[8] * p.z + transl[2];
284            p_proj.x = p_transf.x / p_transf.z;
285            p_proj.y = p_transf.y / p_transf.z;
286            if (norm(p_proj - image_normalized.at<Point2f>(0, i)) < max_dist)
287                inliers->push_back(i);
288        }
289    }
290}
291
292#endif
293