OpenCV使用RANSAC的仿射变换估计 estimateAffine2D

OpenCV自带有findHomography这个用RANSAC随机采样求透视变换的方法,很好用，但是没有一个类似的求仿射的。

自带的getAffineTransform只是简单的使用三对点。

而estimateAffine3D使用的是三维坐标,转换起来有点不方便，而且我在使用中发现,即使把z坐标设置为0，有时候求出来的模型竟然100%都是内点，OpenCV的源码,自己提取,封装了一下.用的是SVN的Trunk,主版本2.32

有几个改动:

1.OpenCV的estimator都是继承自CvModelEstimator2,而这个父类并不是导出类,所以只能把代码都再写一遍

2.据我观察,估计时内部用的是64位浮点数,增加计算精度,我把getAffineTransform也再写了一遍,对应64位精度

//Affine2D.hpp

class Affine2DEstimator
{
public:
 Affine2DEstimator();
 int runKernel( const CvMat* m1, const CvMat* m2, CvMat* model );
 bool runRANSAC( const CvMat* m1, const CvMat* m2, CvMat* model,
  CvMat* mask, double threshold,
  double confidence=0.99, int maxIters=2000 );
 bool getSubset( const CvMat* m1, const CvMat* m2,
  CvMat* ms1, CvMat* ms2, int maxAttempts=1000 );
 bool checkSubset( const CvMat* ms1, int count );
 int findInliers( const CvMat* m1, const CvMat* m2,
  const CvMat* model, CvMat* error,
  CvMat* mask, double threshold );
 void computeReprojError( const CvMat* m1, const CvMat* m2, const CvMat* model, CvMat* error );
protected:
 CvRNG rng;
 int modelPoints;
 CvSize modelSize;
 int maxBasicSolutions;
 bool checkPartialSubsets;
};

int estimateAffine2D(cv::InputArray _from, cv::InputArray _to,
 cv::OutputArray _out, cv::OutputArray _inliers,
 double param1=3, double param2=0.99);

--------------------------------------------------------------------------

int Affine2DEstimator::findInliers( const CvMat* m1, const CvMat* m2,
 const CvMat* model, CvMat* _err,
 CvMat* _mask, double threshold )
{
 int i, count = _err->rows*_err->cols, goodCount = 0;
 const float* err = _err->data.fl;
 uchar* mask = _mask->data.ptr;

 computeReprojError( m1, m2, model, _err );
 threshold *= threshold;
 for( i = 0; i < count; i++ )
  goodCount += mask[i] = err[i] <= threshold;
 return goodCount;
}

void Affine2DEstimator::computeReprojError( const CvMat* m1, const CvMat* m2, const CvMat* model, CvMat* error )
{
 int count = m1->rows * m1->cols;
 const CvPoint2D64f* from = reinterpret_cast<const CvPoint2D64f*>(m1->data.ptr);
 const CvPoint2D64f* to  = reinterpret_cast<const CvPoint2D64f*>(m2->data.ptr);   
 const double* F = model->data.db;
 float* err = error->data.fl;

 for(int i = 0; i < count; i++ )
 {
  const CvPoint2D64f& f = from[i];
  const CvPoint2D64f& t = to[i];

  double a = F[0]*f.x + F[1]*f.y + F[2] - t.x;
  double b = F[3]*f.x + F[4]*f.y + F[5] - t.y;
 
  err[i] = (float)sqrt(a*a + b*b);     
 }
}

bool Affine2DEstimator::runRANSAC( const CvMat* m1, const CvMat* m2, CvMat* model,
 CvMat* mask0, double reprojThreshold,
 double confidence, int maxIters )
{
 bool result = false;
 cv::Ptr<CvMat> mask = cvCloneMat(mask0);
 cv::Ptr<CvMat> models, err, tmask;
 cv::Ptr<CvMat> ms1, ms2;

 int iter, niters = maxIters;
 int count = m1->rows*m1->cols, maxGoodCount = 0;
 CV_Assert( CV_ARE_SIZES_EQ(m1, m2) && CV_ARE_SIZES_EQ(m1, mask) );

 if( count < modelPoints )
  return false;

 models = cvCreateMat( modelSize.height*maxBasicSolutions, modelSize.width, CV_64FC1 );
 err = cvCreateMat( 1, count, CV_32FC1 );
 tmask = cvCreateMat( 1, count, CV_8UC1 );

 if( count > modelPoints )
 {
  ms1 = cvCreateMat( 1, modelPoints, m1->type );
  ms2 = cvCreateMat( 1, modelPoints, m2->type );
 }
 else
 {
  niters = 1;
  ms1 = cvCloneMat(m1);
  ms2 = cvCloneMat(m2);
 }

 for( iter = 0; iter < niters; iter++ )
 {
  int i, goodCount, nmodels;
  if( count > modelPoints )
  {
   bool found = getSubset( m1, m2, ms1, ms2, 300 );
   if( !found )
   {
    if( iter == 0 )
     return false;
    break;
   }
  }

  nmodels = runKernel( ms1, ms2, models );
  if( nmodels <= 0 )
   continue;
  for( i = 0; i < nmodels; i++ )
  {
   CvMat model_i;
   cvGetRows( models, &model_i, i*modelSize.height, (i+1)*modelSize.height );
   goodCount = findInliers( m1, m2, &model_i, err, tmask, reprojThreshold );

   if( goodCount > MAX(maxGoodCount, modelPoints-1) )
   {
    std::swap(tmask, mask);
    cvCopy( &model_i, model );
    maxGoodCount = goodCount;
    niters = cvRANSACUpdateNumIters( confidence,
     (double)(count - goodCount)/count, modelPoints, niters );
   }
  }
 }

 if( maxGoodCount > 0 )
 {
  if( mask != mask0 )
   cvCopy( mask, mask0 );
  result = true;
 }

 return result;
}

Mat getAffineTransform64f( const Point2d src[], const Point2d dst[] )
{
 Mat M(2, 3, CV_64F), X(6, 1, CV_64F, M.data);
 double a[6*6], b[6];
 Mat A(6, 6, CV_64F, a), B(6, 1, CV_64F, b);

 for( int i = 0; i < 3; i++ )
 {
  int j = i*12;
  int k = i*12+6;
  a[j] = a[k+3] = src[i].x;
  a[j+1] = a[k+4] = src[i].y;
  a[j+2] = a[k+5] = 1;
  a[j+3] = a[j+4] = a[j+5] = 0;
  a[k] = a[k+1] = a[k+2] = 0;
  b[i*2] = dst[i].x;
  b[i*2+1] = dst[i].y;
 }

 solve( A, B, X );
 return M;
}

int Affine2DEstimator::runKernel( const CvMat* m1, const CvMat* m2, CvMat* model )
{ 
 const Point2d* from = reinterpret_cast<const Point2d*>(m1->data.ptr);
 const Point2d* to  = reinterpret_cast<const Point2d*>(m2->data.ptr);
 Mat M0 = cv::cvarrToMat(model);
 Mat M=getAffineTransform64f(from,to);
 CV_Assert( M.size() == M0.size() );
 M.convertTo(M0, M0.type());

 return model!=NULL?1:0;
}

int estimateAffine2D(InputArray _from, InputArray _to,
 OutputArray _out, OutputArray _inliers,
 double param1, double param2)
{
 Mat from = _from.getMat(), to = _to.getMat();
 int count = from.checkVector(2, CV_32F);

 CV_Assert( count >= 0 && to.checkVector(2, CV_32F) == count );

 _out.create(2, 3, CV_64F);
 Mat out = _out.getMat();

 _inliers.create(count, 1, CV_8U, -1, true);
 Mat inliers = _inliers.getMat();
 inliers = Scalar::all(1);

 Mat dFrom, dTo;
 from.convertTo(dFrom, CV_64F);
 to.convertTo(dTo, CV_64F);

 CvMat F2x3 = out;
 CvMat mask  = inliers;
 CvMat m1 = dFrom;
 CvMat m2 = dTo;

 const double epsilon = numeric_limits<double>::epsilon();       
 param1 = param1 <= 0 ? 3 : param1;
 param2 = (param2 < epsilon) ? 0.99 : (param2 > 1 - epsilon) ? 0.99 : param2;

 return Affine2DEstimator().runRANSAC(&m1, &m2, &F2x3, &mask, param1, param2 );   
}

bool Affine2DEstimator::getSubset( const CvMat* m1, const CvMat* m2,
 CvMat* ms1, CvMat* ms2, int maxAttempts )
{
 cv::AutoBuffer<int> _idx(modelPoints);
 int* idx = _idx;
 int i = 0, j, k, idx_i, iters = 0;
 int type = CV_MAT_TYPE(m1->type), elemSize = CV_ELEM_SIZE(type);
 const int *m1ptr = m1->data.i, *m2ptr = m2->data.i;
 int *ms1ptr = ms1->data.i, *ms2ptr = ms2->data.i;
 int count = m1->cols*m1->rows;

 assert( CV_IS_MAT_CONT(m1->type & m2->type) && (elemSize % sizeof(int) == 0) );
 elemSize /= sizeof(int);

 for(; iters < maxAttempts; iters++)
 {
  for( i = 0; i < modelPoints && iters < maxAttempts; )
  {
   idx[i] = idx_i = cvRandInt(&rng) % count;
   for( j = 0; j < i; j++ )
    if( idx_i == idx[j] )
     break;
   if( j < i )
    continue;
   for( k = 0; k < elemSize; k++ )
   {
    ms1ptr[i*elemSize + k] = m1ptr[idx_i*elemSize + k];
    ms2ptr[i*elemSize + k] = m2ptr[idx_i*elemSize + k];
   }
   if( checkPartialSubsets && (!checkSubset( ms1, i+1 ) || !checkSubset( ms2, i+1 )))
   {
    iters++;
    continue;
   }
   i++;
  }
  if( !checkPartialSubsets && i == modelPoints &&
   (!checkSubset( ms1, i ) || !checkSubset( ms2, i )))
   continue;
  break;
 }

 return i == modelPoints && iters < maxAttempts;
}

bool Affine2DEstimator::checkSubset( const CvMat* ms1, int count )
{
 int j, k, i, i0, i1;
 CvPoint2D64f* ptr = (CvPoint2D64f*)ms1->data.ptr;

 assert( CV_MAT_TYPE(ms1->type) == CV_64FC2 );

 if( checkPartialSubsets )
  i0 = i1 = count - 1;
 else
  i0 = 0, i1 = count - 1;

 for( i = i0; i <= i1; i++ )
 {
  // check that the i-th selected point does not belong
  // to a line connecting some previously selected points
  for( j = 0; j < i; j++ )
  {
   double dx1 = ptr[j].x - ptr[i].x;
   double dy1 = ptr[j].y - ptr[i].y;
   for( k = 0; k < j; k++ )
   {
    double dx2 = ptr[k].x - ptr[i].x;
    double dy2 = ptr[k].y - ptr[i].y;
    if( fabs(dx2*dy1 - dy2*dx1) <= FLT_EPSILON*(fabs(dx1) + fabs(dy1) + fabs(dx2) + fabs(dy2)))
     break;
   }
   if( k < j )
    break;
  }
  if( j < i )
   break;
 }

 return i >= i1;
}

Affine2DEstimator::Affine2DEstimator() : modelPoints(3),modelSize(cvSize(3, 2)),maxBasicSolutions(1)
{
 checkPartialSubsets = true;
 rng = cvRNG(-1);
}

--------------------------------------分割线 --------------------------------------

--------------------------------------分割线 --------------------------------------

OpenCV的详细介绍：请点这里
OpenCV的下载地址：请点这里