OpenCV C 向量 DMatch 到 C#

std::vector< DMatch > FeatureDetection(char* source, char* itempl, int x, int y ,int width, int height) // Features2D + Homography to find a known object
{
  /// Load image and template
  Mat img_gray = imread( source, CV_LOAD_IMAGE_GRAYSCALE );
  img = img_gray(Rect(x, y, width, height));

  templ = imread( itempl, CV_LOAD_IMAGE_GRAYSCALE );

  //– Step 1: Detect the keypoints using SURF Detector
  int minHessian = 400;

  SurfFeatureDetector detector( minHessian );

  std::vector<KeyPoint> keypoints_1, keypoints_2;

  detector.detect( templ, keypoints_1 );
  detector.detect( img, keypoints_2 );

  //– Step 2: Calculate descriptors (feature vectors)
  SurfDescriptorExtractor extractor;

  Mat descriptors_1, descriptors_2;

  extractor.compute( templ, keypoints_1, descriptors_1 );
  extractor.compute( img, keypoints_2, descriptors_2 );

  //– Step 3: Matching descriptor vectors with a brute force matcher
  BFMatcher matcher(NORM_L2);
  std::vector< DMatch > matches;
  matcher.match( descriptors_1, descriptors_2, matches );

  return matches;
}

[DllImport("OpenCVTest1.dll", CallingConvention = CallingConvention.Cdecl)]
public static extern **???** FeatureDetection(…);

vector<Point2f> matched_points1, matched_points2; // these are your points that match

for (int i=0;i<matches.size();i++)
{
    // this is how the DMatch structure stores the matching information
    int idx1=matches[i].trainIdx;
    int idx2=matches[i].queryIdx;

    //now use those match indices to get the keypoints, add to your two lists of points
    matched_points1.push_back(keypoints_1[idx1].pt);
    matched_points2.push_back(keypoints_2[idx2].pt);
}