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 //                        Intel License Agreement
11 //                For Open Source Computer Vision Library
12 //
13 // Copyright (C) 2000, Intel Corporation, all rights reserved.
14 // Third party copyrights are property of their respective owners.
15 //
16 // Redistribution and use in source and binary forms, with or without modification,
17 // are permitted provided that the following conditions are met:
18 //
19 //   * Redistribution's of source code must retain the above copyright notice,
20 //     this list of conditions and the following disclaimer.
21 //
22 //   * Redistribution's in binary form must reproduce the above copyright notice,
23 //     this list of conditions and the following disclaimer in the documentation
24 //     and/or other materials provided with the distribution.
25 //
26 //   * The name of Intel Corporation may not be used to endorse or promote products
27 //     derived from this software without specific prior written permission.
28 //
29 // This software is provided by the copyright holders and contributors "as is" and
30 // any express or implied warranties, including, but not limited to, the implied
31 // warranties of merchantability and fitness for a particular purpose are disclaimed.
32 // In no event shall the Intel Corporation or contributors be liable for any direct,
33 // indirect, incidental, special, exemplary, or consequential damages
34 // (including, but not limited to, procurement of substitute goods or services;
35 // loss of use, data, or profits; or business interruption) however caused
36 // and on any theory of liability, whether in contract, strict liability,
37 // or tort (including negligence or otherwise) arising in any way out of
38 // the use of this software, even if advised of the possibility of such damage.
39 //
40 //M*/
41 
42 #include "precomp.hpp"
43 #include "opencl_kernels_imgproc.hpp"
44 
45 #include <cstdio>
46 #include <vector>
47 #include <iostream>
48 #include <functional>
49 
50 namespace cv
51 {
52 
53 struct greaterThanPtr :
54         public std::binary_function<const float *, const float *, bool>
55 {
operator ()cv::greaterThanPtr56     bool operator () (const float * a, const float * b) const
57     { return *a > *b; }
58 };
59 
60 #ifdef HAVE_OPENCL
61 
62 struct Corner
63 {
64     float val;
65     short y;
66     short x;
67 
operator <cv::Corner68     bool operator < (const Corner & c) const
69     {  return val > c.val; }
70 };
71 
ocl_goodFeaturesToTrack(InputArray _image,OutputArray _corners,int maxCorners,double qualityLevel,double minDistance,InputArray _mask,int blockSize,bool useHarrisDetector,double harrisK)72 static bool ocl_goodFeaturesToTrack( InputArray _image, OutputArray _corners,
73                                      int maxCorners, double qualityLevel, double minDistance,
74                                      InputArray _mask, int blockSize,
75                                      bool useHarrisDetector, double harrisK )
76 {
77     UMat eig, maxEigenValue;
78     if( useHarrisDetector )
79         cornerHarris( _image, eig, blockSize, 3, harrisK );
80     else
81         cornerMinEigenVal( _image, eig, blockSize, 3 );
82 
83     Size imgsize = _image.size();
84     size_t total, i, j, ncorners = 0, possibleCornersCount =
85             std::max(1024, static_cast<int>(imgsize.area() * 0.1));
86     bool haveMask = !_mask.empty();
87     UMat corners_buffer(1, (int)possibleCornersCount + 1, CV_32FC2);
88     CV_Assert(sizeof(Corner) == corners_buffer.elemSize());
89     Mat tmpCorners;
90 
91     // find threshold
92     {
93         CV_Assert(eig.type() == CV_32FC1);
94         int dbsize = ocl::Device::getDefault().maxComputeUnits();
95         size_t wgs = ocl::Device::getDefault().maxWorkGroupSize();
96 
97         int wgs2_aligned = 1;
98         while (wgs2_aligned < (int)wgs)
99             wgs2_aligned <<= 1;
100         wgs2_aligned >>= 1;
101 
102         ocl::Kernel k("maxEigenVal", ocl::imgproc::gftt_oclsrc,
103                       format("-D OP_MAX_EIGEN_VAL -D WGS=%d -D groupnum=%d -D WGS2_ALIGNED=%d%s",
104                              (int)wgs, dbsize, wgs2_aligned, haveMask ? " -D HAVE_MASK" : ""));
105         if (k.empty())
106             return false;
107 
108         UMat mask = _mask.getUMat();
109         maxEigenValue.create(1, dbsize, CV_32FC1);
110 
111         ocl::KernelArg eigarg = ocl::KernelArg::ReadOnlyNoSize(eig),
112                 dbarg = ocl::KernelArg::PtrWriteOnly(maxEigenValue),
113                 maskarg = ocl::KernelArg::ReadOnlyNoSize(mask),
114                 cornersarg = ocl::KernelArg::PtrWriteOnly(corners_buffer);
115 
116         if (haveMask)
117             k.args(eigarg, eig.cols, (int)eig.total(), dbarg, maskarg);
118         else
119             k.args(eigarg, eig.cols, (int)eig.total(), dbarg);
120 
121         size_t globalsize = dbsize * wgs;
122         if (!k.run(1, &globalsize, &wgs, false))
123             return false;
124 
125         ocl::Kernel k2("maxEigenValTask", ocl::imgproc::gftt_oclsrc,
126                        format("-D OP_MAX_EIGEN_VAL -D WGS=%d -D WGS2_ALIGNED=%d -D groupnum=%d",
127                               wgs, wgs2_aligned, dbsize));
128         if (k2.empty())
129             return false;
130 
131         k2.args(dbarg, (float)qualityLevel, cornersarg);
132 
133         if (!k2.runTask(false))
134             return false;
135     }
136 
137     // collect list of pointers to features - put them into temporary image
138     {
139         ocl::Kernel k("findCorners", ocl::imgproc::gftt_oclsrc,
140                       format("-D OP_FIND_CORNERS%s", haveMask ? " -D HAVE_MASK" : ""));
141         if (k.empty())
142             return false;
143 
144         ocl::KernelArg eigarg = ocl::KernelArg::ReadOnlyNoSize(eig),
145                 cornersarg = ocl::KernelArg::PtrWriteOnly(corners_buffer),
146                 thresholdarg = ocl::KernelArg::PtrReadOnly(maxEigenValue);
147 
148         if (!haveMask)
149             k.args(eigarg, cornersarg, eig.rows - 2, eig.cols - 2, thresholdarg,
150                   (int)possibleCornersCount);
151         else
152         {
153             UMat mask = _mask.getUMat();
154             k.args(eigarg, ocl::KernelArg::ReadOnlyNoSize(mask),
155                    cornersarg, eig.rows - 2, eig.cols - 2,
156                    thresholdarg, (int)possibleCornersCount);
157         }
158 
159         size_t globalsize[2] = { eig.cols - 2, eig.rows - 2 };
160         if (!k.run(2, globalsize, NULL, false))
161             return false;
162 
163         tmpCorners = corners_buffer.getMat(ACCESS_RW);
164         total = std::min<size_t>(tmpCorners.at<Vec2i>(0, 0)[0], possibleCornersCount);
165         if (total == 0)
166         {
167             _corners.release();
168             return true;
169         }
170     }
171 
172     Corner* corner_ptr = tmpCorners.ptr<Corner>() + 1;
173     std::sort(corner_ptr, corner_ptr + total);
174 
175     std::vector<Point2f> corners;
176     corners.reserve(total);
177 
178     if (minDistance >= 1)
179     {
180          // Partition the image into larger grids
181         int w = imgsize.width, h = imgsize.height;
182 
183         const int cell_size = cvRound(minDistance);
184         const int grid_width = (w + cell_size - 1) / cell_size;
185         const int grid_height = (h + cell_size - 1) / cell_size;
186 
187         std::vector<std::vector<Point2f> > grid(grid_width*grid_height);
188         minDistance *= minDistance;
189 
190         for( i = 0; i < total; i++ )
191         {
192             const Corner & c = corner_ptr[i];
193             bool good = true;
194 
195             int x_cell = c.x / cell_size;
196             int y_cell = c.y / cell_size;
197 
198             int x1 = x_cell - 1;
199             int y1 = y_cell - 1;
200             int x2 = x_cell + 1;
201             int y2 = y_cell + 1;
202 
203             // boundary check
204             x1 = std::max(0, x1);
205             y1 = std::max(0, y1);
206             x2 = std::min(grid_width - 1, x2);
207             y2 = std::min(grid_height - 1, y2);
208 
209             for( int yy = y1; yy <= y2; yy++ )
210                 for( int xx = x1; xx <= x2; xx++ )
211                 {
212                     std::vector<Point2f> &m = grid[yy * grid_width + xx];
213 
214                     if( m.size() )
215                     {
216                         for(j = 0; j < m.size(); j++)
217                         {
218                             float dx = c.x - m[j].x;
219                             float dy = c.y - m[j].y;
220 
221                             if( dx*dx + dy*dy < minDistance )
222                             {
223                                 good = false;
224                                 goto break_out;
225                             }
226                         }
227                     }
228                 }
229 
230             break_out:
231 
232             if (good)
233             {
234                 grid[y_cell*grid_width + x_cell].push_back(Point2f((float)c.x, (float)c.y));
235 
236                 corners.push_back(Point2f((float)c.x, (float)c.y));
237                 ++ncorners;
238 
239                 if( maxCorners > 0 && (int)ncorners == maxCorners )
240                     break;
241             }
242         }
243     }
244     else
245     {
246         for( i = 0; i < total; i++ )
247         {
248             const Corner & c = corner_ptr[i];
249 
250             corners.push_back(Point2f((float)c.x, (float)c.y));
251             ++ncorners;
252             if( maxCorners > 0 && (int)ncorners == maxCorners )
253                 break;
254         }
255     }
256 
257     Mat(corners).convertTo(_corners, _corners.fixedType() ? _corners.type() : CV_32F);
258     return true;
259 }
260 
261 #endif
262 
263 }
264 
goodFeaturesToTrack(InputArray _image,OutputArray _corners,int maxCorners,double qualityLevel,double minDistance,InputArray _mask,int blockSize,bool useHarrisDetector,double harrisK)265 void cv::goodFeaturesToTrack( InputArray _image, OutputArray _corners,
266                               int maxCorners, double qualityLevel, double minDistance,
267                               InputArray _mask, int blockSize,
268                               bool useHarrisDetector, double harrisK )
269 {
270     CV_Assert( qualityLevel > 0 && minDistance >= 0 && maxCorners >= 0 );
271     CV_Assert( _mask.empty() || (_mask.type() == CV_8UC1 && _mask.sameSize(_image)) );
272 
273     CV_OCL_RUN(_image.dims() <= 2 && _image.isUMat(),
274                ocl_goodFeaturesToTrack(_image, _corners, maxCorners, qualityLevel, minDistance,
275                                     _mask, blockSize, useHarrisDetector, harrisK))
276 
277     Mat image = _image.getMat(), eig, tmp;
278     if (image.empty())
279     {
280         _corners.release();
281         return;
282     }
283 
284     if( useHarrisDetector )
285         cornerHarris( image, eig, blockSize, 3, harrisK );
286     else
287         cornerMinEigenVal( image, eig, blockSize, 3 );
288 
289     double maxVal = 0;
290     minMaxLoc( eig, 0, &maxVal, 0, 0, _mask );
291     threshold( eig, eig, maxVal*qualityLevel, 0, THRESH_TOZERO );
292     dilate( eig, tmp, Mat());
293 
294     Size imgsize = image.size();
295     std::vector<const float*> tmpCorners;
296 
297     // collect list of pointers to features - put them into temporary image
298     Mat mask = _mask.getMat();
299     for( int y = 1; y < imgsize.height - 1; y++ )
300     {
301         const float* eig_data = (const float*)eig.ptr(y);
302         const float* tmp_data = (const float*)tmp.ptr(y);
303         const uchar* mask_data = mask.data ? mask.ptr(y) : 0;
304 
305         for( int x = 1; x < imgsize.width - 1; x++ )
306         {
307             float val = eig_data[x];
308             if( val != 0 && val == tmp_data[x] && (!mask_data || mask_data[x]) )
309                 tmpCorners.push_back(eig_data + x);
310         }
311     }
312     std::sort( tmpCorners.begin(), tmpCorners.end(), greaterThanPtr() );
313 
314     std::vector<Point2f> corners;
315     size_t i, j, total = tmpCorners.size(), ncorners = 0;
316 
317     if (minDistance >= 1)
318     {
319          // Partition the image into larger grids
320         int w = image.cols;
321         int h = image.rows;
322 
323         const int cell_size = cvRound(minDistance);
324         const int grid_width = (w + cell_size - 1) / cell_size;
325         const int grid_height = (h + cell_size - 1) / cell_size;
326 
327         std::vector<std::vector<Point2f> > grid(grid_width*grid_height);
328 
329         minDistance *= minDistance;
330 
331         for( i = 0; i < total; i++ )
332         {
333             int ofs = (int)((const uchar*)tmpCorners[i] - eig.ptr());
334             int y = (int)(ofs / eig.step);
335             int x = (int)((ofs - y*eig.step)/sizeof(float));
336 
337             bool good = true;
338 
339             int x_cell = x / cell_size;
340             int y_cell = y / cell_size;
341 
342             int x1 = x_cell - 1;
343             int y1 = y_cell - 1;
344             int x2 = x_cell + 1;
345             int y2 = y_cell + 1;
346 
347             // boundary check
348             x1 = std::max(0, x1);
349             y1 = std::max(0, y1);
350             x2 = std::min(grid_width-1, x2);
351             y2 = std::min(grid_height-1, y2);
352 
353             for( int yy = y1; yy <= y2; yy++ )
354                 for( int xx = x1; xx <= x2; xx++ )
355                 {
356                     std::vector <Point2f> &m = grid[yy*grid_width + xx];
357 
358                     if( m.size() )
359                     {
360                         for(j = 0; j < m.size(); j++)
361                         {
362                             float dx = x - m[j].x;
363                             float dy = y - m[j].y;
364 
365                             if( dx*dx + dy*dy < minDistance )
366                             {
367                                 good = false;
368                                 goto break_out;
369                             }
370                         }
371                     }
372                 }
373 
374             break_out:
375 
376             if (good)
377             {
378                 grid[y_cell*grid_width + x_cell].push_back(Point2f((float)x, (float)y));
379 
380                 corners.push_back(Point2f((float)x, (float)y));
381                 ++ncorners;
382 
383                 if( maxCorners > 0 && (int)ncorners == maxCorners )
384                     break;
385             }
386         }
387     }
388     else
389     {
390         for( i = 0; i < total; i++ )
391         {
392             int ofs = (int)((const uchar*)tmpCorners[i] - eig.ptr());
393             int y = (int)(ofs / eig.step);
394             int x = (int)((ofs - y*eig.step)/sizeof(float));
395 
396             corners.push_back(Point2f((float)x, (float)y));
397             ++ncorners;
398             if( maxCorners > 0 && (int)ncorners == maxCorners )
399                 break;
400         }
401     }
402 
403     Mat(corners).convertTo(_corners, _corners.fixedType() ? _corners.type() : CV_32F);
404 }
405 
406 CV_IMPL void
cvGoodFeaturesToTrack(const void * _image,void *,void *,CvPoint2D32f * _corners,int * _corner_count,double quality_level,double min_distance,const void * _maskImage,int block_size,int use_harris,double harris_k)407 cvGoodFeaturesToTrack( const void* _image, void*, void*,
408                        CvPoint2D32f* _corners, int *_corner_count,
409                        double quality_level, double min_distance,
410                        const void* _maskImage, int block_size,
411                        int use_harris, double harris_k )
412 {
413     cv::Mat image = cv::cvarrToMat(_image), mask;
414     std::vector<cv::Point2f> corners;
415 
416     if( _maskImage )
417         mask = cv::cvarrToMat(_maskImage);
418 
419     CV_Assert( _corners && _corner_count );
420     cv::goodFeaturesToTrack( image, corners, *_corner_count, quality_level,
421         min_distance, mask, block_size, use_harris != 0, harris_k );
422 
423     size_t i, ncorners = corners.size();
424     for( i = 0; i < ncorners; i++ )
425         _corners[i] = corners[i];
426     *_corner_count = (int)ncorners;
427 }
428 
429 /* End of file. */
430