vtk Registration (二 ) vtkIterativeClosestPointTransform 点云数据配准最经典的方法是迭代最近点算法 ICP

   ICP

点云配准分为两步，先做粗配准，再做精配准：

• 粗配准（Coarse Global Registeration）：基于局部几何特征
• 精配准（Fine Local Registeration）：需要初始位姿（initial alignment）

  这里讲一下 VTK 的 ICP IterativeClosestPointTransform,可以用于精配准；

精配准见 ​​vtkLandmarkTransform​​

点云数据配准最经典的方法是迭代最近点算法（Iterative Closest Points,ICP）。ICP算法是一个迭代的过程，每次迭代中对于源数据点P找到目标点集Q中的最近点，然后给予最小二乘原理求解当前的变换矩阵T。通过不断迭代迭代直至收敛，即完成了点集的配准;

vtkSmartPointer<vtkDICOMImageReader>  QtMyShow::Read_VTK_DICOM_DIR(const char * path)
{
  vtkSmartPointer<vtkDICOMImageReader> reader =
    vtkSmartPointer<vtkDICOMImageReader>::New();

  reader->SetDirectoryName(path);
  reader->Update();

  return reader;

}

vtkSmartPointer<vtkPolyData>  QtMyShow::ReadDicom(const char * path)
{
  vtkSmartPointer<vtkDICOMImageReader> reader =
    Read_VTK_DICOM_DIR(path);
 

    //抽取等值面为骨头的信息
  vtkMarchingCubes *boneExtractor = vtkMarchingCubes::New();
  boneExtractor->SetInputData(reader->GetOutput());
  boneExtractor->SetValue(0, 85); //设置提取的等值信息
  boneExtractor->Update();
 
 剔除旧的或废除的数据单元，提高绘制速度
  //vtkStripper *boneStripper = vtkStripper::New(); //三角带连接                                     
  //boneStripper->SetInputData(boneExtractor->GetOutput());
  //boneStripper->Update();

  vtkSmartPointer<vtkPolyData> data = vtkSmartPointer<vtkPolyData>::New();
  data->DeepCopy(boneExtractor->GetOutput());
  return data;
  //return reader->GetOutput();
}

int main()
{
  vtkSmartPointer<vtkVertexGlyphFilter>  sourceGlyph =
    vtkSmartPointer<vtkVertexGlyphFilter>::New();
  //sourceGlyph->SetInputData(ReadDicom("E:/2"));
  sourceGlyph->SetInputData(ReadDicom("E:/2"));
  sourceGlyph->Update();
 
  vtkSmartPointer<vtkTransform> trans =
    vtkSmartPointer<vtkTransform>::New();
  trans->Translate(0.2, 0.1, 0.1);
  trans->RotateX(-30);

  vtkSmartPointer<vtkTransformPolyDataFilter> transformFilter1 =
    vtkSmartPointer<vtkTransformPolyDataFilter>::New();
  transformFilter1->SetInputData(sourceGlyph->GetOutput());
  transformFilter1->SetTransform(trans);
  transformFilter1->Update();
 


  target ->DeepCopy(sourceGlyph->GetOutput());
  source->DeepCopy(transformFilter1->GetOutput());

    vtkNew<vtkIterativeClosestPointTransform> icp;
  icp->SetSource(source);
  icp->SetTarget(target);
  icp->GetLandmarkTransform()->SetModeToRigidBody();
  
  icp->SetMaximumNumberOfIterations(50);
  icp->StartByMatchingCentroidsOn();
  icp->Modified();
  icp->Update();

 
  // Get the resulting transformation matrix (this matrix takes the source
  // points to the target points)
  vtkSmartPointer<vtkMatrix4x4> m = icp->GetMatrix();
  std::cout << "The resulting matrix is: " << *m << std::endl;

  // Transform the source points by the ICP solution
  vtkNew<vtkTransformPolyDataFilter> icpTransformFilter;
  icpTransformFilter->SetInputData(source);
  icpTransformFilter->SetTransform(icp);
  icpTransformFilter->Update();
}

我测试了一下，对于同源的数据可能效果不错，不过对于异源数据，多模数据，效果很差；