Drawing matches from Nearest Neighbour Distance Ratio

Question

In openCV I have adapted this tutorial code in my application

http://docs.opencv.org/2.4.2/doc/tutorials/features2d/feature_homography/feature_homography.html#feature-homography

I have been trying to prune the matches using the Nearest Neighbour Distance Ratio to only draw matches above a certain threshold value.

  double ratio = 0.9;
  std::vector< vector<DMatch > > nnMatches;
  std::vector< DMatch > good_NNmatches;
  matcher.knnMatch(descriptors_scene, descriptors_object, nnMatches, 2 );

  for(int k = 0; k < nnMatches.size(); k++)
  {
      if(nnMatches[k][0].distance / nnMatches[k][1].distance > ratio)
      {
          good_NNmatches.push_back(nnMatches[k][0]);
      }
  }

I am then trying to draw the matches in good_NNmatches using the same method demonstrated in the tutorial but I get the following error:

OpenCV Error: Assertion failed (i1 >= 0 && i1 < static_cast<int>(keypoints1.size())) in     drawMatches, file /Users/cgray/Downloads/opencv-2.4.6/modules/features2d/src/draw.cpp, line 207
libc++abi.dylib: terminating with uncaught exception of type cv::Exception: /Users/cgray/Downloads/opencv-2.4.6/modules/features2d/src/draw.cpp:207: error: (-215) i1 >= 0 && i1 < static_cast<int>(keypoints1.size()) in function drawMatches

When trying to call drawMatches using good_nnMatches instead of good_matches as described in the tutorial.

drawMatches( roiImg, keypoints_object, compare, keypoints_scene,
                     good_NNmatches, img_matches, Scalar::all(-1), Scalar::all(-1),
                     vector<char>(), DrawMatchesFlags::NOT_DRAW_SINGLE_POINTS );

I don't know if that is the cause of your problem, but your conditionnal is wrong. You should add a match in `good_NNmatches` if `nnMatches[k][0].distance / nnMatches[k][1].distance < ratio`. — BConic, Apr 18 '14 at 11:37
That didn't solve the assertion error but I have changed the condition now and found a fix to the issue, Thanks. — Tom smith, Apr 18 '14 at 11:42

adinutzyc21 · Answer 1 · 2014-06-17T15:03:19.213

I just had the same problem. In my case I was doing

cv::Mat output;
drawMatches( smallimg, keyptsSm, bigimg, keyptsBg, matchesBig2Small, output );

Where the matches matchesBig2Small were computed from bigimg to smallimg

std::vector<cv::DMatch> matchesBig2Small; 
//match from bigimg to smallimg

So the error you got is actually saying that it's looking for the keypoint index in smallimg that actually refers to the second image, which is bigger so you're exceeding the size of smallimg. So either change the call to drawMatches to:

drawMatches( bigimg, keyptsBg, smallimg, keyptsSm, matchesBig2Small, output );

Or change the way you're computing the matches to be from smallimg to bigimg, so instead of matchesBig2Small you'd have matchesSmall2Big, and then the call would be:

std::vector<cv::DMatch> matchesSmall2Big; 
//match from smallimg to bigimg
drawMatches( smallimg, keyptsSm, bigimg, keyptsBg, matchesSmall2Big, output );

Thanks, this was precisely my mistake. – Avio Jan 13 '17 at 17:07 — Avio, Jan 13 '17 at 17:07

score 0 · Accepted Answer · answered Apr 18 '14 at 11:38

Ok after some trial and error I appear to have solved this issue now.

I have changed the drawMatches method to swap the "roiImg" and key points with "compare" and key points.

drawMatches(compare, keypoints_scene, roiImg, keypoints_object,
               good_NNmatches, img_matches, Scalar::all(-1), Scalar::all(-1),
               vector<char>(), DrawMatchesFlags::NOT_DRAW_SINGLE_POINTS );

This stopped the assertion error. But due to swapping the values around when it came to retrieving the key points I also had to make some changes. trainIdx and queryIdx have also been swapped to account for the previous changes made.

obj.push_back( keypoints_object[ good_NNmatches[i].trainIdx ].pt );
scene.push_back( keypoints_scene[ good_NNmatches[i].queryIdx ].pt );

score 0 · Answer 3 · answered Apr 15 '18 at 21:38

I have done it and i don not have anyproblems I hope that this code can help you

double NNDR;

    for(int i=0;i<(int)ORB_ORB_matches.size(); i++){

        NNDR= ( ORB_ORB_matches[i][0].distance/ORB_ORB_matches[i][1].distance );

        if(NNDR <= 0.9){

            ORB_single_matches.push_back (ORB_ORB_matches[i][0]);

        }

    }
drawMatches( image_1, keypoints_1, image_2, keypoints_2, ORB_single_matches, img_matches,Scalar::all(-1),Scalar::all(-1),vector<char>(),`enter code here`          DrawMatchesFlags::DRAW_RICH_KEYPOINTS|DrawMatchesFlags::NOT_DRAW_SINGLE_POINTS);

Drawing matches from Nearest Neighbour Distance Ratio

3 Answers3