• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 "_cv.h"
43 
icvCompressPoints(T * ptr,const uchar * mask,int mstep,int count)44 template<typename T> int icvCompressPoints( T* ptr, const uchar* mask, int mstep, int count )
45 {
46     int i, j;
47     for( i = j = 0; i < count; i++ )
48         if( mask[i*mstep] )
49         {
50             if( i > j )
51                 ptr[j] = ptr[i];
52             j++;
53         }
54     return j;
55 }
56 
57 class CvModelEstimator2
58 {
59 public:
60     CvModelEstimator2(int _modelPoints, CvSize _modelSize, int _maxBasicSolutions);
61     virtual ~CvModelEstimator2();
62 
63     virtual int runKernel( const CvMat* m1, const CvMat* m2, CvMat* model )=0;
64     virtual bool runLMeDS( const CvMat* m1, const CvMat* m2, CvMat* model,
65                            CvMat* mask, double confidence=0.99, int maxIters=1000 );
66     virtual bool runRANSAC( const CvMat* m1, const CvMat* m2, CvMat* model,
67                             CvMat* mask, double threshold,
68                             double confidence=0.99, int maxIters=1000 );
refine(const CvMat *,const CvMat *,CvMat *,int)69     virtual bool refine( const CvMat*, const CvMat*, CvMat*, int ) { return true; }
70     virtual void setSeed( int64 seed );
71 
72 protected:
73     virtual void computeReprojError( const CvMat* m1, const CvMat* m2,
74                                      const CvMat* model, CvMat* error ) = 0;
75     virtual int findInliers( const CvMat* m1, const CvMat* m2,
76                              const CvMat* model, CvMat* error,
77                              CvMat* mask, double threshold );
78     virtual bool getSubset( const CvMat* m1, const CvMat* m2,
79                             CvMat* ms1, CvMat* ms2, int maxAttempts=1000 );
80     virtual bool checkSubset( const CvMat* ms1, int count );
81 
82     CvRNG rng;
83     int modelPoints;
84     CvSize modelSize;
85     int maxBasicSolutions;
86     bool checkPartialSubsets;
87 };
88 
89 
CvModelEstimator2(int _modelPoints,CvSize _modelSize,int _maxBasicSolutions)90 CvModelEstimator2::CvModelEstimator2(int _modelPoints, CvSize _modelSize, int _maxBasicSolutions)
91 {
92     modelPoints = _modelPoints;
93     modelSize = _modelSize;
94     maxBasicSolutions = _maxBasicSolutions;
95     checkPartialSubsets = true;
96     rng = cvRNG(-1);
97 }
98 
~CvModelEstimator2()99 CvModelEstimator2::~CvModelEstimator2()
100 {
101 }
102 
setSeed(int64 seed)103 void CvModelEstimator2::setSeed( int64 seed )
104 {
105     rng = cvRNG(seed);
106 }
107 
108 
findInliers(const CvMat * m1,const CvMat * m2,const CvMat * model,CvMat * _err,CvMat * _mask,double threshold)109 int CvModelEstimator2::findInliers( const CvMat* m1, const CvMat* m2,
110                                     const CvMat* model, CvMat* _err,
111                                     CvMat* _mask, double threshold )
112 {
113     int i, count = _err->rows*_err->cols, goodCount = 0;
114     const float* err = _err->data.fl;
115     uchar* mask = _mask->data.ptr;
116 
117     computeReprojError( m1, m2, model, _err );
118     threshold *= threshold;
119     for( i = 0; i < count; i++ )
120         goodCount += mask[i] = err[i] <= threshold;
121     return goodCount;
122 }
123 
124 
125 CV_IMPL int
cvRANSACUpdateNumIters(double p,double ep,int model_points,int max_iters)126 cvRANSACUpdateNumIters( double p, double ep,
127                         int model_points, int max_iters )
128 {
129     int result = 0;
130 
131     CV_FUNCNAME( "cvRANSACUpdateNumIters" );
132 
133     __BEGIN__;
134 
135     double num, denom;
136 
137     if( model_points <= 0 )
138         CV_ERROR( CV_StsOutOfRange, "the number of model points should be positive" );
139 
140     p = MAX(p, 0.);
141     p = MIN(p, 1.);
142     ep = MAX(ep, 0.);
143     ep = MIN(ep, 1.);
144 
145     // avoid inf's & nan's
146     num = MAX(1. - p, DBL_MIN);
147     denom = 1. - pow(1. - ep,model_points);
148     if( denom < DBL_MIN )
149         EXIT;
150 
151     num = log(num);
152     denom = log(denom);
153 
154     result = denom >= 0 || -num >= max_iters*(-denom) ?
155         max_iters : cvRound(num/denom);
156 
157     __END__;
158 
159     return result;
160 }
161 
runRANSAC(const CvMat * m1,const CvMat * m2,CvMat * model,CvMat * mask,double reprojThreshold,double confidence,int maxIters)162 bool CvModelEstimator2::runRANSAC( const CvMat* m1, const CvMat* m2, CvMat* model,
163                                         CvMat* mask, double reprojThreshold,
164                                         double confidence, int maxIters )
165 {
166     bool result = false;
167     CvMat* mask0 = mask, *tmask = 0, *t;
168     CvMat* models = 0, *err = 0;
169     CvMat *ms1 = 0, *ms2 = 0;
170 
171     CV_FUNCNAME( "CvModelEstimator2::estimateRansac" );
172 
173     __BEGIN__;
174 
175     int iter, niters = maxIters;
176     int count = m1->rows*m1->cols, maxGoodCount = 0;
177     CV_ASSERT( CV_ARE_SIZES_EQ(m1, m2) && CV_ARE_SIZES_EQ(m1, mask) );
178 
179     if( count < modelPoints )
180         EXIT;
181 
182     models = cvCreateMat( modelSize.height*maxBasicSolutions, modelSize.width, CV_64FC1 );
183     err = cvCreateMat( 1, count, CV_32FC1 );
184     tmask = cvCreateMat( 1, count, CV_8UC1 );
185 
186     if( count > modelPoints )
187     {
188         ms1 = cvCreateMat( 1, modelPoints, m1->type );
189         ms2 = cvCreateMat( 1, modelPoints, m2->type );
190     }
191     else
192     {
193         niters = 1;
194         ms1 = (CvMat*)m1;
195         ms2 = (CvMat*)m2;
196     }
197 
198     for( iter = 0; iter < niters; iter++ )
199     {
200         int i, goodCount, nmodels;
201         if( count > modelPoints )
202         {
203             bool found = getSubset( m1, m2, ms1, ms2, modelPoints );
204             if( !found )
205             {
206                 if( iter == 0 )
207                     EXIT;
208                 break;
209             }
210         }
211 
212         nmodels = runKernel( ms1, ms2, models );
213         if( nmodels <= 0 )
214             continue;
215         for( i = 0; i < nmodels; i++ )
216         {
217             CvMat model_i;
218             cvGetRows( models, &model_i, i*modelSize.height, (i+1)*modelSize.height );
219             goodCount = findInliers( m1, m2, &model_i, err, tmask, reprojThreshold );
220 
221             if( goodCount > MAX(maxGoodCount, modelPoints-1) )
222             {
223                 CV_SWAP( tmask, mask, t );
224                 cvCopy( &model_i, model );
225                 maxGoodCount = goodCount;
226                 niters = cvRANSACUpdateNumIters( confidence,
227                     (double)(count - goodCount)/count, modelPoints, niters );
228             }
229         }
230     }
231 
232     if( maxGoodCount > 0 )
233     {
234         if( mask != mask0 )
235         {
236             CV_SWAP( tmask, mask, t );
237             cvCopy( tmask, mask );
238         }
239         result = true;
240     }
241 
242     __END__;
243 
244     if( ms1 != m1 )
245         cvReleaseMat( &ms1 );
246     if( ms2 != m2 )
247         cvReleaseMat( &ms2 );
248     cvReleaseMat( &models );
249     cvReleaseMat( &err );
250     cvReleaseMat( &tmask );
251     return result;
252 }
253 
254 
CV_IMPLEMENT_QSORT(icvSortDistances,int,CV_LT)255 static CV_IMPLEMENT_QSORT( icvSortDistances, int, CV_LT )
256 
257 bool CvModelEstimator2::runLMeDS( const CvMat* m1, const CvMat* m2, CvMat* model,
258                                   CvMat* mask, double confidence, int maxIters )
259 {
260     const double outlierRatio = 0.45;
261     bool result = false;
262     CvMat* models = 0;
263     CvMat *ms1 = 0, *ms2 = 0;
264     CvMat* err = 0;
265 
266     CV_FUNCNAME( "CvModelEstimator2::estimateLMeDS" );
267 
268     __BEGIN__;
269 
270     int iter, niters = maxIters;
271     int count = m1->rows*m1->cols;
272     double minMedian = DBL_MAX, sigma;
273 
274     CV_ASSERT( CV_ARE_SIZES_EQ(m1, m2) && CV_ARE_SIZES_EQ(m1, mask) );
275 
276     if( count < modelPoints )
277         EXIT;
278 
279     models = cvCreateMat( modelSize.height*maxBasicSolutions, modelSize.width, CV_64FC1 );
280     err = cvCreateMat( 1, count, CV_32FC1 );
281 
282     if( count > modelPoints )
283     {
284         ms1 = cvCreateMat( 1, modelPoints, m1->type );
285         ms2 = cvCreateMat( 1, modelPoints, m2->type );
286     }
287     else
288     {
289         niters = 1;
290         ms1 = (CvMat*)m1;
291         ms2 = (CvMat*)m2;
292     }
293 
294     niters = cvRound(log(1-confidence)/log(1-pow(1-outlierRatio,(double)modelPoints)));
295     niters = MIN( MAX(niters, 3), maxIters );
296 
297     for( iter = 0; iter < niters; iter++ )
298     {
299         int i, nmodels;
300         if( count > modelPoints )
301         {
302             bool found = getSubset( m1, m2, ms1, ms2, 300 );
303             if( !found )
304             {
305                 if( iter == 0 )
306                     EXIT;
307                 break;
308             }
309         }
310 
311         nmodels = runKernel( ms1, ms2, models );
312         if( nmodels <= 0 )
313             continue;
314         for( i = 0; i < nmodels; i++ )
315         {
316             CvMat model_i;
317             cvGetRows( models, &model_i, i*modelSize.height, (i+1)*modelSize.height );
318             computeReprojError( m1, m2, &model_i, err );
319             icvSortDistances( err->data.i, count, 0 );
320 
321             double median = count % 2 != 0 ?
322                 err->data.fl[count/2] : (err->data.fl[count/2-1] + err->data.fl[count/2])*0.5;
323 
324             if( median < minMedian )
325             {
326                 minMedian = median;
327                 cvCopy( &model_i, model );
328             }
329         }
330     }
331 
332     if( minMedian < DBL_MAX )
333     {
334         sigma = 2.5*1.4826*(1 + 5./(count - modelPoints))*sqrt(minMedian);
335         sigma = MAX( sigma, FLT_EPSILON*100 );
336 
337         count = findInliers( m1, m2, model, err, mask, sigma );
338         result = count >= modelPoints;
339     }
340 
341     __END__;
342 
343     if( ms1 != m1 )
344         cvReleaseMat( &ms1 );
345     if( ms2 != m2 )
346         cvReleaseMat( &ms2 );
347     cvReleaseMat( &models );
348     cvReleaseMat( &err );
349     return result;
350 }
351 
352 
getSubset(const CvMat * m1,const CvMat * m2,CvMat * ms1,CvMat * ms2,int maxAttempts)353 bool CvModelEstimator2::getSubset( const CvMat* m1, const CvMat* m2,
354                                    CvMat* ms1, CvMat* ms2, int maxAttempts )
355 {
356     int* idx = (int*)cvStackAlloc( modelPoints*sizeof(idx[0]) );
357     int i, j, k, idx_i, iters = 0;
358     int type = CV_MAT_TYPE(m1->type), elemSize = CV_ELEM_SIZE(type);
359     const int *m1ptr = m1->data.i, *m2ptr = m2->data.i;
360     int *ms1ptr = ms1->data.i, *ms2ptr = ms2->data.i;
361     int count = m1->cols*m1->rows;
362 
363     assert( CV_IS_MAT_CONT(m1->type & m2->type) && (elemSize % sizeof(int) == 0) );
364     elemSize /= sizeof(int);
365 
366     for(;;)
367     {
368         for( i = 0; i < modelPoints && iters < maxAttempts; iters++ )
369         {
370             idx[i] = idx_i = cvRandInt(&rng) % count;
371             for( j = 0; j < i; j++ )
372                 if( idx_i == idx[j] )
373                     break;
374             if( j < i )
375                 continue;
376             for( k = 0; k < elemSize; k++ )
377             {
378                 ms1ptr[i*elemSize + k] = m1ptr[idx_i*elemSize + k];
379                 ms2ptr[i*elemSize + k] = m2ptr[idx_i*elemSize + k];
380             }
381             if( checkPartialSubsets && (!checkSubset( ms1, i+1 ) || !checkSubset( ms2, i+1 )))
382                 continue;
383             i++;
384             iters = 0;
385         }
386         if( !checkPartialSubsets && i == modelPoints &&
387             (!checkSubset( ms1, i+1 ) || !checkSubset( ms2, i+1 )))
388             continue;
389         break;
390     }
391 
392     return i == modelPoints;
393 }
394 
395 
checkSubset(const CvMat * m,int count)396 bool CvModelEstimator2::checkSubset( const CvMat* m, int count )
397 {
398     int j, k, i = count-1;
399     CvPoint2D64f* ptr = (CvPoint2D64f*)m->data.ptr;
400 
401     assert( CV_MAT_TYPE(m->type) == CV_64FC2 );
402 
403     // check that the i-th selected point does not belong
404     // to a line connecting some previously selected points
405     for( j = 0; j < i; j++ )
406     {
407         double dx1 = ptr[j].x - ptr[i].x;
408         double dy1 = ptr[j].y - ptr[i].y;
409         for( k = 0; k < j; k++ )
410         {
411             double dx2 = ptr[k].x - ptr[i].x;
412             double dy2 = ptr[k].y - ptr[i].y;
413             if( fabs(dx2*dy1 - dy2*dx1) < FLT_EPSILON*(fabs(dx1) + fabs(dy1) + fabs(dx2) + fabs(dy2)))
414                 break;
415         }
416         if( k < j )
417             break;
418     }
419 
420     return j == i;
421 }
422 
423 
424 class CvHomographyEstimator : public CvModelEstimator2
425 {
426 public:
427     CvHomographyEstimator( int modelPoints );
428 
429     virtual int runKernel( const CvMat* m1, const CvMat* m2, CvMat* model );
430     virtual bool refine( const CvMat* m1, const CvMat* m2,
431                          CvMat* model, int maxIters );
432 protected:
433     virtual void computeReprojError( const CvMat* m1, const CvMat* m2,
434                                      const CvMat* model, CvMat* error );
435 };
436 
437 
CvHomographyEstimator(int _modelPoints)438 CvHomographyEstimator::CvHomographyEstimator(int _modelPoints)
439     : CvModelEstimator2(_modelPoints, cvSize(3,3), 1)
440 {
441     assert( _modelPoints == 4 || _modelPoints == 5 );
442 }
443 
runKernel(const CvMat * m1,const CvMat * m2,CvMat * H)444 int CvHomographyEstimator::runKernel( const CvMat* m1, const CvMat* m2, CvMat* H )
445 {
446     int i, count = m1->rows*m1->cols;
447     const CvPoint2D64f* M = (const CvPoint2D64f*)m1->data.ptr;
448     const CvPoint2D64f* m = (const CvPoint2D64f*)m2->data.ptr;
449 
450     double LtL[9][9], W[9][9], V[9][9];
451     CvMat _LtL = cvMat( 9, 9, CV_64F, LtL );
452     CvMat _W = cvMat( 9, 9, CV_64F, W );
453     CvMat _V = cvMat( 9, 9, CV_64F, V );
454     CvMat _H0 = cvMat( 3, 3, CV_64F, V[8] );
455     CvMat _Htemp = cvMat( 3, 3, CV_64F, V[7] );
456     CvPoint2D64f cM={0,0}, cm={0,0}, sM={0,0}, sm={0,0};
457 
458     for( i = 0; i < count; i++ )
459     {
460         cm.x += m[i].x; cm.y += m[i].y;
461         cM.x += M[i].x; cM.y += M[i].y;
462     }
463 
464     cm.x /= count; cm.y /= count;
465     cM.x /= count; cM.y /= count;
466 
467     for( i = 0; i < count; i++ )
468     {
469         sm.x += fabs(m[i].x - cm.x);
470         sm.y += fabs(m[i].y - cm.y);
471         sM.x += fabs(M[i].x - cM.x);
472         sM.y += fabs(M[i].y - cM.y);
473     }
474 
475     sm.x = count/sm.x; sm.y = count/sm.y;
476     sM.x = count/sM.x; sM.y = count/sM.y;
477 
478     double invHnorm[9] = { 1./sm.x, 0, cm.x, 0, 1./sm.y, cm.y, 0, 0, 1 };
479     double Hnorm2[9] = { sM.x, 0, -cM.x*sM.x, 0, sM.y, -cM.y*sM.y, 0, 0, 1 };
480     CvMat _invHnorm = cvMat( 3, 3, CV_64FC1, invHnorm );
481     CvMat _Hnorm2 = cvMat( 3, 3, CV_64FC1, Hnorm2 );
482 
483     cvZero( &_LtL );
484     for( i = 0; i < count; i++ )
485     {
486         double x = (m[i].x - cm.x)*sm.x, y = (m[i].y - cm.y)*sm.y;
487         double X = (M[i].x - cM.x)*sM.x, Y = (M[i].y - cM.y)*sM.y;
488         double Lx[] = { X, Y, 1, 0, 0, 0, -x*X, -x*Y, -x };
489         double Ly[] = { 0, 0, 0, X, Y, 1, -y*X, -y*Y, -y };
490         int j, k;
491         for( j = 0; j < 9; j++ )
492             for( k = j; k < 9; k++ )
493                 LtL[j][k] += Lx[j]*Lx[k] + Ly[j]*Ly[k];
494     }
495     cvCompleteSymm( &_LtL );
496 
497     cvSVD( &_LtL, &_W, 0, &_V, CV_SVD_MODIFY_A + CV_SVD_V_T );
498     cvMatMul( &_invHnorm, &_H0, &_Htemp );
499     cvMatMul( &_Htemp, &_Hnorm2, &_H0 );
500     cvConvertScale( &_H0, H, 1./_H0.data.db[8] );
501 
502     return 1;
503 }
504 
505 
computeReprojError(const CvMat * m1,const CvMat * m2,const CvMat * model,CvMat * _err)506 void CvHomographyEstimator::computeReprojError( const CvMat* m1, const CvMat* m2,
507                                                 const CvMat* model, CvMat* _err )
508 {
509     int i, count = m1->rows*m1->cols;
510     const CvPoint2D64f* M = (const CvPoint2D64f*)m1->data.ptr;
511     const CvPoint2D64f* m = (const CvPoint2D64f*)m2->data.ptr;
512     const double* H = model->data.db;
513     float* err = _err->data.fl;
514 
515     for( i = 0; i < count; i++ )
516     {
517         double ww = 1./(H[6]*M[i].x + H[7]*M[i].y + 1.);
518         double dx = (H[0]*M[i].x + H[1]*M[i].y + H[2])*ww - m[i].x;
519         double dy = (H[3]*M[i].x + H[4]*M[i].y + H[5])*ww - m[i].y;
520         err[i] = (float)(dx*dx + dy*dy);
521     }
522 }
523 
refine(const CvMat * m1,const CvMat * m2,CvMat * model,int maxIters)524 bool CvHomographyEstimator::refine( const CvMat* m1, const CvMat* m2, CvMat* model, int maxIters )
525 {
526     CvLevMarq solver(8, 0, cvTermCriteria(CV_TERMCRIT_ITER+CV_TERMCRIT_EPS, maxIters, DBL_EPSILON));
527     int i, j, k, count = m1->rows*m1->cols;
528     const CvPoint2D64f* M = (const CvPoint2D64f*)m1->data.ptr;
529     const CvPoint2D64f* m = (const CvPoint2D64f*)m2->data.ptr;
530     CvMat modelPart = cvMat( solver.param->rows, solver.param->cols, model->type, model->data.ptr );
531     cvCopy( &modelPart, solver.param );
532 
533     for(;;)
534     {
535         const CvMat* _param = 0;
536         CvMat *_JtJ = 0, *_JtErr = 0;
537         double* _errNorm = 0;
538 
539         if( !solver.updateAlt( _param, _JtJ, _JtErr, _errNorm ))
540             break;
541 
542         for( i = 0; i < count; i++ )
543         {
544             const double* h = _param->data.db;
545             double Mx = M[i].x, My = M[i].y;
546             double ww = 1./(h[6]*Mx + h[7]*My + 1.);
547             double _xi = (h[0]*Mx + h[1]*My + h[2])*ww;
548             double _yi = (h[3]*Mx + h[4]*My + h[5])*ww;
549             double err[] = { _xi - m[i].x, _yi - m[i].y };
550             if( _JtJ || _JtErr )
551             {
552                 double J[][8] =
553                 {
554                     { Mx*ww, My*ww, ww, 0, 0, 0, -Mx*ww*_xi, -My*ww*_xi },
555                     { 0, 0, 0, Mx*ww, My*ww, ww, -Mx*ww*_yi, -My*ww*_yi }
556                 };
557 
558                 for( j = 0; j < 8; j++ )
559                 {
560                     for( k = j; k < 8; k++ )
561                         _JtJ->data.db[j*8+k] += J[0][j]*J[0][k] + J[1][j]*J[1][k];
562                     _JtErr->data.db[j] += J[0][j]*err[0] + J[1][j]*err[1];
563                 }
564             }
565             if( _errNorm )
566                 *_errNorm += err[0]*err[0] + err[1]*err[1];
567         }
568     }
569 
570     cvCopy( solver.param, &modelPart );
571     return true;
572 }
573 
574 
575 CV_IMPL int
cvFindHomography(const CvMat * objectPoints,const CvMat * imagePoints,CvMat * __H,int method,double ransacReprojThreshold,CvMat * mask)576 cvFindHomography( const CvMat* objectPoints, const CvMat* imagePoints,
577                   CvMat* __H, int method, double ransacReprojThreshold,
578                   CvMat* mask )
579 {
580     const double confidence = 0.99;
581     bool result = false;
582     CvMat *m = 0, *M = 0, *tempMask = 0;
583 
584     CV_FUNCNAME( "cvFindHomography" );
585 
586     __BEGIN__;
587 
588     double H[9];
589     CvMat _H = cvMat( 3, 3, CV_64FC1, H );
590     int count;
591 
592     CV_ASSERT( CV_IS_MAT(imagePoints) && CV_IS_MAT(objectPoints) );
593 
594     count = MAX(imagePoints->cols, imagePoints->rows);
595     CV_ASSERT( count >= 4 );
596 
597     m = cvCreateMat( 1, count, CV_64FC2 );
598     cvConvertPointsHomogeneous( imagePoints, m );
599 
600     M = cvCreateMat( 1, count, CV_64FC2 );
601     cvConvertPointsHomogeneous( objectPoints, M );
602 
603     if( mask )
604     {
605         CV_ASSERT( CV_IS_MASK_ARR(mask) && CV_IS_MAT_CONT(mask->type) &&
606             (mask->rows == 1 || mask->cols == 1) &&
607             mask->rows*mask->cols == count );
608         tempMask = mask;
609     }
610     else if( count > 4 )
611         tempMask = cvCreateMat( 1, count, CV_8U );
612     if( tempMask )
613         cvSet( tempMask, cvScalarAll(1.) );
614 
615     {
616     CvHomographyEstimator estimator( MIN(count, 5) );
617     if( count == 4 )
618         method = 0;
619     if( method == CV_LMEDS )
620         result = estimator.runLMeDS( M, m, &_H, tempMask, confidence );
621     else if( method == CV_RANSAC )
622         result = estimator.runRANSAC( M, m, &_H, tempMask, ransacReprojThreshold, confidence );
623     else
624         result = estimator.runKernel( M, m, &_H ) > 0;
625 
626     if( result && count > 4 )
627     {
628         icvCompressPoints( (CvPoint2D64f*)M->data.ptr, tempMask->data.ptr, 1, count );
629         count = icvCompressPoints( (CvPoint2D64f*)m->data.ptr, tempMask->data.ptr, 1, count );
630         M->cols = m->cols = count;
631         estimator.refine( M, m, &_H, 10 );
632     }
633     }
634 
635     if( result )
636         cvConvert( &_H, __H );
637 
638     __END__;
639 
640     cvReleaseMat( &m );
641     cvReleaseMat( &M );
642     if( tempMask != mask )
643         cvReleaseMat( &tempMask );
644 
645     return (int)result;
646 }
647 
648 
649 /* Evaluation of Fundamental Matrix from point correspondences.
650    The original code has been written by Valery Mosyagin */
651 
652 /* The algorithms (except for RANSAC) and the notation have been taken from
653    Zhengyou Zhang's research report
654    "Determining the Epipolar Geometry and its Uncertainty: A Review"
655    that can be found at http://www-sop.inria.fr/robotvis/personnel/zzhang/zzhang-eng.html */
656 
657 /************************************** 7-point algorithm *******************************/
658 class CvFMEstimator : public CvModelEstimator2
659 {
660 public:
661     CvFMEstimator( int _modelPoints );
662 
663     virtual int runKernel( const CvMat* m1, const CvMat* m2, CvMat* model );
664     virtual int run7Point( const CvMat* m1, const CvMat* m2, CvMat* model );
665     virtual int run8Point( const CvMat* m1, const CvMat* m2, CvMat* model );
666 protected:
667     virtual void computeReprojError( const CvMat* m1, const CvMat* m2,
668                                      const CvMat* model, CvMat* error );
669 };
670 
CvFMEstimator(int _modelPoints)671 CvFMEstimator::CvFMEstimator( int _modelPoints )
672 : CvModelEstimator2( _modelPoints, cvSize(3,3), _modelPoints == 7 ? 3 : 1 )
673 {
674     assert( _modelPoints == 7 || _modelPoints == 8 );
675 }
676 
677 
runKernel(const CvMat * m1,const CvMat * m2,CvMat * model)678 int CvFMEstimator::runKernel( const CvMat* m1, const CvMat* m2, CvMat* model )
679 {
680     return modelPoints == 7 ? run7Point( m1, m2, model ) : run8Point( m1, m2, model );
681 }
682 
run7Point(const CvMat * _m1,const CvMat * _m2,CvMat * _fmatrix)683 int CvFMEstimator::run7Point( const CvMat* _m1, const CvMat* _m2, CvMat* _fmatrix )
684 {
685     double a[7*9], w[7], v[9*9], c[4], r[3];
686     double* f1, *f2;
687     double t0, t1, t2;
688     CvMat A = cvMat( 7, 9, CV_64F, a );
689     CvMat V = cvMat( 9, 9, CV_64F, v );
690     CvMat W = cvMat( 7, 1, CV_64F, w );
691     CvMat coeffs = cvMat( 1, 4, CV_64F, c );
692     CvMat roots = cvMat( 1, 3, CV_64F, r );
693     const CvPoint2D64f* m1 = (const CvPoint2D64f*)_m1->data.ptr;
694     const CvPoint2D64f* m2 = (const CvPoint2D64f*)_m2->data.ptr;
695     double* fmatrix = _fmatrix->data.db;
696     int i, k, n;
697 
698     // form a linear system: i-th row of A(=a) represents
699     // the equation: (m2[i], 1)'*F*(m1[i], 1) = 0
700     for( i = 0; i < 7; i++ )
701     {
702         double x0 = m1[i].x, y0 = m1[i].y;
703         double x1 = m2[i].x, y1 = m2[i].y;
704 
705         a[i*9+0] = x1*x0;
706         a[i*9+1] = x1*y0;
707         a[i*9+2] = x1;
708         a[i*9+3] = y1*x0;
709         a[i*9+4] = y1*y0;
710         a[i*9+5] = y1;
711         a[i*9+6] = x0;
712         a[i*9+7] = y0;
713         a[i*9+8] = 1;
714     }
715 
716     // A*(f11 f12 ... f33)' = 0 is singular (7 equations for 9 variables), so
717     // the solution is linear subspace of dimensionality 2.
718     // => use the last two singular vectors as a basis of the space
719     // (according to SVD properties)
720     cvSVD( &A, &W, 0, &V, CV_SVD_MODIFY_A + CV_SVD_V_T );
721     f1 = v + 7*9;
722     f2 = v + 8*9;
723 
724     // f1, f2 is a basis => lambda*f1 + mu*f2 is an arbitrary f. matrix.
725     // as it is determined up to a scale, normalize lambda & mu (lambda + mu = 1),
726     // so f ~ lambda*f1 + (1 - lambda)*f2.
727     // use the additional constraint det(f) = det(lambda*f1 + (1-lambda)*f2) to find lambda.
728     // it will be a cubic equation.
729     // find c - polynomial coefficients.
730     for( i = 0; i < 9; i++ )
731         f1[i] -= f2[i];
732 
733     t0 = f2[4]*f2[8] - f2[5]*f2[7];
734     t1 = f2[3]*f2[8] - f2[5]*f2[6];
735     t2 = f2[3]*f2[7] - f2[4]*f2[6];
736 
737     c[3] = f2[0]*t0 - f2[1]*t1 + f2[2]*t2;
738 
739     c[2] = f1[0]*t0 - f1[1]*t1 + f1[2]*t2 -
740            f1[3]*(f2[1]*f2[8] - f2[2]*f2[7]) +
741            f1[4]*(f2[0]*f2[8] - f2[2]*f2[6]) -
742            f1[5]*(f2[0]*f2[7] - f2[1]*f2[6]) +
743            f1[6]*(f2[1]*f2[5] - f2[2]*f2[4]) -
744            f1[7]*(f2[0]*f2[5] - f2[2]*f2[3]) +
745            f1[8]*(f2[0]*f2[4] - f2[1]*f2[3]);
746 
747     t0 = f1[4]*f1[8] - f1[5]*f1[7];
748     t1 = f1[3]*f1[8] - f1[5]*f1[6];
749     t2 = f1[3]*f1[7] - f1[4]*f1[6];
750 
751     c[1] = f2[0]*t0 - f2[1]*t1 + f2[2]*t2 -
752            f2[3]*(f1[1]*f1[8] - f1[2]*f1[7]) +
753            f2[4]*(f1[0]*f1[8] - f1[2]*f1[6]) -
754            f2[5]*(f1[0]*f1[7] - f1[1]*f1[6]) +
755            f2[6]*(f1[1]*f1[5] - f1[2]*f1[4]) -
756            f2[7]*(f1[0]*f1[5] - f1[2]*f1[3]) +
757            f2[8]*(f1[0]*f1[4] - f1[1]*f1[3]);
758 
759     c[0] = f1[0]*t0 - f1[1]*t1 + f1[2]*t2;
760 
761     // solve the cubic equation; there can be 1 to 3 roots ...
762     n = cvSolveCubic( &coeffs, &roots );
763 
764     if( n < 1 || n > 3 )
765         return n;
766 
767     for( k = 0; k < n; k++, fmatrix += 9 )
768     {
769         // for each root form the fundamental matrix
770         double lambda = r[k], mu = 1.;
771         double s = f1[8]*r[k] + f2[8];
772 
773         // normalize each matrix, so that F(3,3) (~fmatrix[8]) == 1
774         if( fabs(s) > DBL_EPSILON )
775         {
776             mu = 1./s;
777             lambda *= mu;
778             fmatrix[8] = 1.;
779         }
780         else
781             fmatrix[8] = 0.;
782 
783         for( i = 0; i < 8; i++ )
784             fmatrix[i] = f1[i]*lambda + f2[i]*mu;
785     }
786 
787     return n;
788 }
789 
790 
run8Point(const CvMat * _m1,const CvMat * _m2,CvMat * _fmatrix)791 int CvFMEstimator::run8Point( const CvMat* _m1, const CvMat* _m2, CvMat* _fmatrix )
792 {
793     double a[9*9], w[9], v[9*9];
794     CvMat W = cvMat( 1, 9, CV_64F, w );
795     CvMat V = cvMat( 9, 9, CV_64F, v );
796     CvMat A = cvMat( 9, 9, CV_64F, a );
797     CvMat U, F0, TF;
798 
799     CvPoint2D64f m0c = {0,0}, m1c = {0,0};
800     double t, scale0 = 0, scale1 = 0;
801 
802     const CvPoint2D64f* m1 = (const CvPoint2D64f*)_m1->data.ptr;
803     const CvPoint2D64f* m2 = (const CvPoint2D64f*)_m2->data.ptr;
804     double* fmatrix = _fmatrix->data.db;
805     int i, j, k, count = _m1->cols*_m1->rows;
806 
807     // compute centers and average distances for each of the two point sets
808     for( i = 0; i < count; i++ )
809     {
810         double x = m1[i].x, y = m1[i].y;
811         m0c.x += x; m0c.y += y;
812 
813         x = m2[i].x, y = m2[i].y;
814         m1c.x += x; m1c.y += y;
815     }
816 
817     // calculate the normalizing transformations for each of the point sets:
818     // after the transformation each set will have the mass center at the coordinate origin
819     // and the average distance from the origin will be ~sqrt(2).
820     t = 1./count;
821     m0c.x *= t; m0c.y *= t;
822     m1c.x *= t; m1c.y *= t;
823 
824     for( i = 0; i < count; i++ )
825     {
826         double x = m1[i].x - m0c.x, y = m1[i].y - m0c.y;
827         scale0 += sqrt(x*x + y*y);
828 
829         x = fabs(m2[i].x - m1c.x), y = fabs(m2[i].y - m1c.y);
830         scale1 += sqrt(x*x + y*y);
831     }
832 
833     scale0 *= t;
834     scale1 *= t;
835 
836     if( scale0 < FLT_EPSILON || scale1 < FLT_EPSILON )
837         return 0;
838 
839     scale0 = sqrt(2.)/scale0;
840     scale1 = sqrt(2.)/scale1;
841 
842     cvZero( &A );
843 
844     // form a linear system Ax=0: for each selected pair of points m1 & m2,
845     // the row of A(=a) represents the coefficients of equation: (m2, 1)'*F*(m1, 1) = 0
846     // to save computation time, we compute (At*A) instead of A and then solve (At*A)x=0.
847     for( i = 0; i < count; i++ )
848     {
849         double x0 = (m1[i].x - m0c.x)*scale0;
850         double y0 = (m1[i].y - m0c.y)*scale0;
851         double x1 = (m2[i].x - m1c.x)*scale1;
852         double y1 = (m2[i].y - m1c.y)*scale1;
853         double r[9] = { x1*x0, x1*y0, x1, y1*x0, y1*y0, y1, x0, y0, 1 };
854         for( j = 0; j < 9; j++ )
855             for( k = 0; k < 9; k++ )
856                 a[j*9+k] += r[j]*r[k];
857     }
858 
859     cvSVD( &A, &W, 0, &V, CV_SVD_MODIFY_A + CV_SVD_V_T );
860 
861     for( i = 0; i < 8; i++ )
862     {
863         if( fabs(w[i]) < DBL_EPSILON )
864             break;
865     }
866 
867     if( i < 7 )
868         return 0;
869 
870     F0 = cvMat( 3, 3, CV_64F, v + 9*8 ); // take the last column of v as a solution of Af = 0
871 
872     // make F0 singular (of rank 2) by decomposing it with SVD,
873     // zeroing the last diagonal element of W and then composing the matrices back.
874 
875     // use v as a temporary storage for different 3x3 matrices
876     W = U = V = TF = F0;
877     W.data.db = v;
878     U.data.db = v + 9;
879     V.data.db = v + 18;
880     TF.data.db = v + 27;
881 
882     cvSVD( &F0, &W, &U, &V, CV_SVD_MODIFY_A + CV_SVD_U_T + CV_SVD_V_T );
883     W.data.db[8] = 0.;
884 
885     // F0 <- U*diag([W(1), W(2), 0])*V'
886     cvGEMM( &U, &W, 1., 0, 0., &TF, CV_GEMM_A_T );
887     cvGEMM( &TF, &V, 1., 0, 0., &F0, 0/*CV_GEMM_B_T*/ );
888 
889     // apply the transformation that is inverse
890     // to what we used to normalize the point coordinates
891     {
892         double tt0[] = { scale0, 0, -scale0*m0c.x, 0, scale0, -scale0*m0c.y, 0, 0, 1 };
893         double tt1[] = { scale1, 0, -scale1*m1c.x, 0, scale1, -scale1*m1c.y, 0, 0, 1 };
894         CvMat T0, T1;
895         T0 = T1 = F0;
896         T0.data.db = tt0;
897         T1.data.db = tt1;
898 
899         // F0 <- T1'*F0*T0
900         cvGEMM( &T1, &F0, 1., 0, 0., &TF, CV_GEMM_A_T );
901         F0.data.db = fmatrix;
902         cvGEMM( &TF, &T0, 1., 0, 0., &F0, 0 );
903 
904         // make F(3,3) = 1
905         if( fabs(F0.data.db[8]) > FLT_EPSILON )
906             cvScale( &F0, &F0, 1./F0.data.db[8] );
907     }
908 
909     return 1;
910 }
911 
912 
computeReprojError(const CvMat * _m1,const CvMat * _m2,const CvMat * model,CvMat * _err)913 void CvFMEstimator::computeReprojError( const CvMat* _m1, const CvMat* _m2,
914                                         const CvMat* model, CvMat* _err )
915 {
916     int i, count = _m1->rows*_m1->cols;
917     const CvPoint2D64f* m1 = (const CvPoint2D64f*)_m1->data.ptr;
918     const CvPoint2D64f* m2 = (const CvPoint2D64f*)_m2->data.ptr;
919     const double* F = model->data.db;
920     float* err = _err->data.fl;
921 
922     for( i = 0; i < count; i++ )
923     {
924         double a, b, c, d1, d2, s1, s2;
925 
926         a = F[0]*m1[i].x + F[1]*m1[i].y + F[2];
927         b = F[3]*m1[i].x + F[4]*m1[i].y + F[5];
928         c = F[6]*m1[i].x + F[7]*m1[i].y + F[8];
929 
930         s2 = 1./(a*a + b*b);
931         d2 = m2[i].x*a + m2[i].y*b + c;
932 
933         a = F[0]*m2[i].x + F[3]*m2[i].y + F[6];
934         b = F[1]*m2[i].x + F[4]*m2[i].y + F[7];
935         c = F[2]*m2[i].x + F[5]*m2[i].y + F[8];
936 
937         s1 = 1./(a*a + b*b);
938         d1 = m1[i].x*a + m1[i].y*b + c;
939 
940         err[i] = (float)(d1*d1*s1 + d2*d2*s2);
941     }
942 }
943 
944 
945 CV_IMPL int
cvFindFundamentalMat(const CvMat * points1,const CvMat * points2,CvMat * fmatrix,int method,double param1,double param2,CvMat * mask)946 cvFindFundamentalMat( const CvMat* points1, const CvMat* points2,
947                       CvMat* fmatrix, int method,
948                       double param1, double param2, CvMat* mask )
949 {
950     int result = 0;
951     CvMat *m1 = 0, *m2 = 0, *tempMask = 0;
952 
953     CV_FUNCNAME( "cvFindFundamentalMat" );
954 
955     __BEGIN__;
956 
957     double F[3*9];
958     CvMat _F3x3 = cvMat( 3, 3, CV_64FC1, F ), _F9x3 = cvMat( 9, 3, CV_64FC1, F );
959     int count;
960 
961     CV_ASSERT( CV_IS_MAT(points1) && CV_IS_MAT(points2) && CV_ARE_SIZES_EQ(points1, points2) );
962     CV_ASSERT( CV_IS_MAT(fmatrix) && fmatrix->cols == 3 &&
963         (fmatrix->rows == 3 || (fmatrix->rows == 9 && method == CV_FM_7POINT)) );
964 
965     count = MAX(points1->cols, points1->rows);
966     if( count < 7 )
967         EXIT;
968 
969     m1 = cvCreateMat( 1, count, CV_64FC2 );
970     cvConvertPointsHomogeneous( points1, m1 );
971 
972     m2 = cvCreateMat( 1, count, CV_64FC2 );
973     cvConvertPointsHomogeneous( points2, m2 );
974 
975     if( mask )
976     {
977         CV_ASSERT( CV_IS_MASK_ARR(mask) && CV_IS_MAT_CONT(mask->type) &&
978             (mask->rows == 1 || mask->cols == 1) &&
979             mask->rows*mask->cols == count );
980         tempMask = mask;
981     }
982     else if( count > 8 )
983         tempMask = cvCreateMat( 1, count, CV_8U );
984     if( tempMask )
985         cvSet( tempMask, cvScalarAll(1.) );
986 
987     {
988     CvFMEstimator estimator( MIN(count, (method & 3) == CV_FM_7POINT ? 7 : 8) );
989     if( count == 7 )
990         result = estimator.run7Point(m1, m2, &_F9x3);
991     else if( count == 8 || method == CV_FM_8POINT )
992         result = estimator.run8Point(m1, m2, &_F3x3);
993     else if( count > 8 )
994     {
995         if( param1 <= 0 )
996             param1 = 3;
997         if( param2 < DBL_EPSILON || param2 > 1 - DBL_EPSILON )
998             param2 = 0.99;
999 
1000         if( (method & ~3) == CV_RANSAC )
1001             result = estimator.runRANSAC(m1, m2, &_F3x3, tempMask, param1, param2 );
1002         else
1003             result = estimator.runLMeDS(m1, m2, &_F3x3, tempMask, param2 );
1004         if( result <= 0 )
1005             EXIT;
1006         icvCompressPoints( (CvPoint2D64f*)m1->data.ptr, tempMask->data.ptr, 1, count );
1007         count = icvCompressPoints( (CvPoint2D64f*)m2->data.ptr, tempMask->data.ptr, 1, count );
1008         assert( count >= 8 );
1009         m1->cols = m2->cols = count;
1010         estimator.run8Point(m1, m2, &_F3x3);
1011     }
1012     }
1013 
1014     if( result )
1015         cvConvert( fmatrix->rows == 3 ? &_F3x3 : &_F9x3, fmatrix );
1016 
1017     __END__;
1018 
1019     cvReleaseMat( &m1 );
1020     cvReleaseMat( &m2 );
1021     if( tempMask != mask )
1022         cvReleaseMat( &tempMask );
1023 
1024     return result;
1025 }
1026 
1027 
1028 CV_IMPL void
cvComputeCorrespondEpilines(const CvMat * points,int pointImageID,const CvMat * fmatrix,CvMat * lines)1029 cvComputeCorrespondEpilines( const CvMat* points, int pointImageID,
1030                              const CvMat* fmatrix, CvMat* lines )
1031 {
1032     CV_FUNCNAME( "cvComputeCorrespondEpilines" );
1033 
1034     __BEGIN__;
1035 
1036     int abc_stride, abc_plane_stride, abc_elem_size;
1037     int plane_stride, stride, elem_size;
1038     int i, dims, count, depth, cn, abc_dims, abc_count, abc_depth, abc_cn;
1039     uchar *ap, *bp, *cp;
1040     const uchar *xp, *yp, *zp;
1041     double f[9];
1042     CvMat F = cvMat( 3, 3, CV_64F, f );
1043 
1044     if( !CV_IS_MAT(points) )
1045         CV_ERROR( !points ? CV_StsNullPtr : CV_StsBadArg, "points parameter is not a valid matrix" );
1046 
1047     depth = CV_MAT_DEPTH(points->type);
1048     cn = CV_MAT_CN(points->type);
1049     if( (depth != CV_32F && depth != CV_64F) || (cn != 1 && cn != 2 && cn != 3) )
1050         CV_ERROR( CV_StsUnsupportedFormat, "The format of point matrix is unsupported" );
1051 
1052     if( points->rows > points->cols )
1053     {
1054         dims = cn*points->cols;
1055         count = points->rows;
1056     }
1057     else
1058     {
1059         if( (points->rows > 1 && cn > 1) || (points->rows == 1 && cn == 1) )
1060             CV_ERROR( CV_StsBadSize, "The point matrix does not have a proper layout (2xn, 3xn, nx2 or nx3)" );
1061         dims = cn * points->rows;
1062         count = points->cols;
1063     }
1064 
1065     if( dims != 2 && dims != 3 )
1066         CV_ERROR( CV_StsOutOfRange, "The dimensionality of points must be 2 or 3" );
1067 
1068     if( !CV_IS_MAT(fmatrix) )
1069         CV_ERROR( !fmatrix ? CV_StsNullPtr : CV_StsBadArg, "fmatrix is not a valid matrix" );
1070 
1071     if( CV_MAT_TYPE(fmatrix->type) != CV_32FC1 && CV_MAT_TYPE(fmatrix->type) != CV_64FC1 )
1072         CV_ERROR( CV_StsUnsupportedFormat, "fundamental matrix must have 32fC1 or 64fC1 type" );
1073 
1074     if( fmatrix->cols != 3 || fmatrix->rows != 3 )
1075         CV_ERROR( CV_StsBadSize, "fundamental matrix must be 3x3" );
1076 
1077     if( !CV_IS_MAT(lines) )
1078         CV_ERROR( !lines ? CV_StsNullPtr : CV_StsBadArg, "lines parameter is not a valid matrix" );
1079 
1080     abc_depth = CV_MAT_DEPTH(lines->type);
1081     abc_cn = CV_MAT_CN(lines->type);
1082     if( (abc_depth != CV_32F && abc_depth != CV_64F) || (abc_cn != 1 && abc_cn != 3) )
1083         CV_ERROR( CV_StsUnsupportedFormat, "The format of the matrix of lines is unsupported" );
1084 
1085     if( lines->rows > lines->cols )
1086     {
1087         abc_dims = abc_cn*lines->cols;
1088         abc_count = lines->rows;
1089     }
1090     else
1091     {
1092         if( (lines->rows > 1 && abc_cn > 1) || (lines->rows == 1 && abc_cn == 1) )
1093             CV_ERROR( CV_StsBadSize, "The lines matrix does not have a proper layout (3xn or nx3)" );
1094         abc_dims = abc_cn * lines->rows;
1095         abc_count = lines->cols;
1096     }
1097 
1098     if( abc_dims != 3 )
1099         CV_ERROR( CV_StsOutOfRange, "The lines matrix does not have a proper layout (3xn or nx3)" );
1100 
1101     if( abc_count != count )
1102         CV_ERROR( CV_StsUnmatchedSizes, "The numbers of points and lines are different" );
1103 
1104     elem_size = CV_ELEM_SIZE(depth);
1105     abc_elem_size = CV_ELEM_SIZE(abc_depth);
1106 
1107     if( points->rows == dims )
1108     {
1109         plane_stride = points->step;
1110         stride = elem_size;
1111     }
1112     else
1113     {
1114         plane_stride = elem_size;
1115         stride = points->rows == 1 ? dims*elem_size : points->step;
1116     }
1117 
1118     if( lines->rows == 3 )
1119     {
1120         abc_plane_stride = lines->step;
1121         abc_stride = abc_elem_size;
1122     }
1123     else
1124     {
1125         abc_plane_stride = abc_elem_size;
1126         abc_stride = lines->rows == 1 ? 3*abc_elem_size : lines->step;
1127     }
1128 
1129     CV_CALL( cvConvert( fmatrix, &F ));
1130     if( pointImageID == 2 )
1131         cvTranspose( &F, &F );
1132 
1133     xp = points->data.ptr;
1134     yp = xp + plane_stride;
1135     zp = dims == 3 ? yp + plane_stride : 0;
1136 
1137     ap = lines->data.ptr;
1138     bp = ap + abc_plane_stride;
1139     cp = bp + abc_plane_stride;
1140 
1141     for( i = 0; i < count; i++ )
1142     {
1143         double x, y, z = 1.;
1144         double a, b, c, nu;
1145 
1146         if( depth == CV_32F )
1147         {
1148             x = *(float*)xp; y = *(float*)yp;
1149             if( zp )
1150                 z = *(float*)zp, zp += stride;
1151         }
1152         else
1153         {
1154             x = *(double*)xp; y = *(double*)yp;
1155             if( zp )
1156                 z = *(double*)zp, zp += stride;
1157         }
1158 
1159         xp += stride; yp += stride;
1160 
1161         a = f[0]*x + f[1]*y + f[2]*z;
1162         b = f[3]*x + f[4]*y + f[5]*z;
1163         c = f[6]*x + f[7]*y + f[8]*z;
1164         nu = a*a + b*b;
1165         nu = nu ? 1./sqrt(nu) : 1.;
1166         a *= nu; b *= nu; c *= nu;
1167 
1168         if( abc_depth == CV_32F )
1169         {
1170             *(float*)ap = (float)a;
1171             *(float*)bp = (float)b;
1172             *(float*)cp = (float)c;
1173         }
1174         else
1175         {
1176             *(double*)ap = a;
1177             *(double*)bp = b;
1178             *(double*)cp = c;
1179         }
1180 
1181         ap += abc_stride;
1182         bp += abc_stride;
1183         cp += abc_stride;
1184     }
1185 
1186     __END__;
1187 }
1188 
1189 
1190 CV_IMPL void
cvConvertPointsHomogeneous(const CvMat * src,CvMat * dst)1191 cvConvertPointsHomogeneous( const CvMat* src, CvMat* dst )
1192 {
1193     CvMat* temp = 0;
1194     CvMat* denom = 0;
1195 
1196     CV_FUNCNAME( "cvConvertPointsHomogeneous" );
1197 
1198     __BEGIN__;
1199 
1200     int i, s_count, s_dims, d_count, d_dims;
1201     CvMat _src, _dst, _ones;
1202     CvMat* ones = 0;
1203 
1204     if( !CV_IS_MAT(src) )
1205         CV_ERROR( !src ? CV_StsNullPtr : CV_StsBadArg,
1206         "The input parameter is not a valid matrix" );
1207 
1208     if( !CV_IS_MAT(dst) )
1209         CV_ERROR( !dst ? CV_StsNullPtr : CV_StsBadArg,
1210         "The output parameter is not a valid matrix" );
1211 
1212     if( src == dst || src->data.ptr == dst->data.ptr )
1213     {
1214         if( src != dst && (!CV_ARE_TYPES_EQ(src, dst) || !CV_ARE_SIZES_EQ(src,dst)) )
1215             CV_ERROR( CV_StsBadArg, "Invalid inplace operation" );
1216         EXIT;
1217     }
1218 
1219     if( src->rows > src->cols )
1220     {
1221         if( !((src->cols > 1) ^ (CV_MAT_CN(src->type) > 1)) )
1222             CV_ERROR( CV_StsBadSize, "Either the number of channels or columns or rows must be =1" );
1223 
1224         s_dims = CV_MAT_CN(src->type)*src->cols;
1225         s_count = src->rows;
1226     }
1227     else
1228     {
1229         if( !((src->rows > 1) ^ (CV_MAT_CN(src->type) > 1)) )
1230             CV_ERROR( CV_StsBadSize, "Either the number of channels or columns or rows must be =1" );
1231 
1232         s_dims = CV_MAT_CN(src->type)*src->rows;
1233         s_count = src->cols;
1234     }
1235 
1236     if( src->rows == 1 || src->cols == 1 )
1237         src = cvReshape( src, &_src, 1, s_count );
1238 
1239     if( dst->rows > dst->cols )
1240     {
1241         if( !((dst->cols > 1) ^ (CV_MAT_CN(dst->type) > 1)) )
1242             CV_ERROR( CV_StsBadSize,
1243             "Either the number of channels or columns or rows in the input matrix must be =1" );
1244 
1245         d_dims = CV_MAT_CN(dst->type)*dst->cols;
1246         d_count = dst->rows;
1247     }
1248     else
1249     {
1250         if( !((dst->rows > 1) ^ (CV_MAT_CN(dst->type) > 1)) )
1251             CV_ERROR( CV_StsBadSize,
1252             "Either the number of channels or columns or rows in the output matrix must be =1" );
1253 
1254         d_dims = CV_MAT_CN(dst->type)*dst->rows;
1255         d_count = dst->cols;
1256     }
1257 
1258     if( dst->rows == 1 || dst->cols == 1 )
1259         dst = cvReshape( dst, &_dst, 1, d_count );
1260 
1261     if( s_count != d_count )
1262         CV_ERROR( CV_StsUnmatchedSizes, "Both matrices must have the same number of points" );
1263 
1264     if( CV_MAT_DEPTH(src->type) < CV_32F || CV_MAT_DEPTH(dst->type) < CV_32F )
1265         CV_ERROR( CV_StsUnsupportedFormat,
1266         "Both matrices must be floating-point (single or double precision)" );
1267 
1268     if( s_dims < 2 || s_dims > 4 || d_dims < 2 || d_dims > 4 )
1269         CV_ERROR( CV_StsOutOfRange,
1270         "Both input and output point dimensionality must be 2, 3 or 4" );
1271 
1272     if( s_dims < d_dims - 1 || s_dims > d_dims + 1 )
1273         CV_ERROR( CV_StsUnmatchedSizes,
1274         "The dimensionalities of input and output point sets differ too much" );
1275 
1276     if( s_dims == d_dims - 1 )
1277     {
1278         if( d_count == dst->rows )
1279         {
1280             ones = cvGetSubRect( dst, &_ones, cvRect( s_dims, 0, 1, d_count ));
1281             dst = cvGetSubRect( dst, &_dst, cvRect( 0, 0, s_dims, d_count ));
1282         }
1283         else
1284         {
1285             ones = cvGetSubRect( dst, &_ones, cvRect( 0, s_dims, d_count, 1 ));
1286             dst = cvGetSubRect( dst, &_dst, cvRect( 0, 0, d_count, s_dims ));
1287         }
1288     }
1289 
1290     if( s_dims <= d_dims )
1291     {
1292         if( src->rows == dst->rows && src->cols == dst->cols )
1293         {
1294             if( CV_ARE_TYPES_EQ( src, dst ) )
1295                 cvCopy( src, dst );
1296             else
1297                 cvConvert( src, dst );
1298         }
1299         else
1300         {
1301             if( !CV_ARE_TYPES_EQ( src, dst ))
1302             {
1303                 CV_CALL( temp = cvCreateMat( src->rows, src->cols, dst->type ));
1304                 cvConvert( src, temp );
1305                 src = temp;
1306             }
1307             cvTranspose( src, dst );
1308         }
1309 
1310         if( ones )
1311             cvSet( ones, cvRealScalar(1.) );
1312     }
1313     else
1314     {
1315         int s_plane_stride, s_stride, d_plane_stride, d_stride, elem_size;
1316 
1317         if( !CV_ARE_TYPES_EQ( src, dst ))
1318         {
1319             CV_CALL( temp = cvCreateMat( src->rows, src->cols, dst->type ));
1320             cvConvert( src, temp );
1321             src = temp;
1322         }
1323 
1324         elem_size = CV_ELEM_SIZE(src->type);
1325 
1326         if( s_count == src->cols )
1327             s_plane_stride = src->step / elem_size, s_stride = 1;
1328         else
1329             s_stride = src->step / elem_size, s_plane_stride = 1;
1330 
1331         if( d_count == dst->cols )
1332             d_plane_stride = dst->step / elem_size, d_stride = 1;
1333         else
1334             d_stride = dst->step / elem_size, d_plane_stride = 1;
1335 
1336         CV_CALL( denom = cvCreateMat( 1, d_count, dst->type ));
1337 
1338         if( CV_MAT_DEPTH(dst->type) == CV_32F )
1339         {
1340             const float* xs = src->data.fl;
1341             const float* ys = xs + s_plane_stride;
1342             const float* zs = 0;
1343             const float* ws = xs + (s_dims - 1)*s_plane_stride;
1344 
1345             float* iw = denom->data.fl;
1346 
1347             float* xd = dst->data.fl;
1348             float* yd = xd + d_plane_stride;
1349             float* zd = 0;
1350 
1351             if( d_dims == 3 )
1352             {
1353                 zs = ys + s_plane_stride;
1354                 zd = yd + d_plane_stride;
1355             }
1356 
1357             for( i = 0; i < d_count; i++, ws += s_stride )
1358             {
1359                 float t = *ws;
1360                 iw[i] = t ? t : 1.f;
1361             }
1362 
1363             cvDiv( 0, denom, denom );
1364 
1365             if( d_dims == 3 )
1366                 for( i = 0; i < d_count; i++ )
1367                 {
1368                     float w = iw[i];
1369                     float x = *xs * w, y = *ys * w, z = *zs * w;
1370                     xs += s_stride; ys += s_stride; zs += s_stride;
1371                     *xd = x; *yd = y; *zd = z;
1372                     xd += d_stride; yd += d_stride; zd += d_stride;
1373                 }
1374             else
1375                 for( i = 0; i < d_count; i++ )
1376                 {
1377                     float w = iw[i];
1378                     float x = *xs * w, y = *ys * w;
1379                     xs += s_stride; ys += s_stride;
1380                     *xd = x; *yd = y;
1381                     xd += d_stride; yd += d_stride;
1382                 }
1383         }
1384         else
1385         {
1386             const double* xs = src->data.db;
1387             const double* ys = xs + s_plane_stride;
1388             const double* zs = 0;
1389             const double* ws = xs + (s_dims - 1)*s_plane_stride;
1390 
1391             double* iw = denom->data.db;
1392 
1393             double* xd = dst->data.db;
1394             double* yd = xd + d_plane_stride;
1395             double* zd = 0;
1396 
1397             if( d_dims == 3 )
1398             {
1399                 zs = ys + s_plane_stride;
1400                 zd = yd + d_plane_stride;
1401             }
1402 
1403             for( i = 0; i < d_count; i++, ws += s_stride )
1404             {
1405                 double t = *ws;
1406                 iw[i] = t ? t : 1.;
1407             }
1408 
1409             cvDiv( 0, denom, denom );
1410 
1411             if( d_dims == 3 )
1412                 for( i = 0; i < d_count; i++ )
1413                 {
1414                     double w = iw[i];
1415                     double x = *xs * w, y = *ys * w, z = *zs * w;
1416                     xs += s_stride; ys += s_stride; zs += s_stride;
1417                     *xd = x; *yd = y; *zd = z;
1418                     xd += d_stride; yd += d_stride; zd += d_stride;
1419                 }
1420             else
1421                 for( i = 0; i < d_count; i++ )
1422                 {
1423                     double w = iw[i];
1424                     double x = *xs * w, y = *ys * w;
1425                     xs += s_stride; ys += s_stride;
1426                     *xd = x; *yd = y;
1427                     xd += d_stride; yd += d_stride;
1428                 }
1429         }
1430     }
1431 
1432     __END__;
1433 
1434     cvReleaseMat( &denom );
1435     cvReleaseMat( &temp );
1436 }
1437 
1438 /* End of file. */
1439