c++ Opencv sift Positiondetection

Question

I work with ubuntu and the Code is in C++ with opencv. I test a little bit to detect some parts of pictures. It works very well, but now I want to find the position in my big picture. Here is the code:

#include...
using namespace cv;

int main(int argc, char** argv) {



Mat img = imread("/home/ubuntu/workspace2/sift/src/inputklein.jpg",CV_LOAD_IMAGE_GRAYSCALE);


 while(1){
             Mat img2 =imread("/home/ubuntu/workspace2/sift/src/input.jpeg",CV_LOAD_IMAGE_GRAYSCALE);   //frame
             //initialize SIFT
             // Create smart pointer for SIFT feature detector.
             SIFT sift;
             vector<KeyPoint> key_points;
             vector<KeyPoint> key_points2;

             //-- Step 1: Detect the keypoints using SURF Detector
             int minHessian = 100;
             SurfFeatureDetector detector( minHessian );
             detector.detect( img, key_points );
             detector.detect( img2, key_points2 );

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

             Mat descriptors1;
             Mat descriptors2;

             extractor.compute( img, key_points, descriptors1 );
             extractor.compute( img2, key_points2, descriptors2 );


             //-- Step 3: Matching descriptor vectors using FLANN matcher
              FlannBasedMatcher matcher;
              std::vector< DMatch > matches;
              matcher.match( descriptors1, descriptors2, matches );

              double max_dist = 20; double min_dist = 10;


              //-- Quick calculation of max and min distances between keypoints
                for( int i = 0; i < descriptors1.rows; i++ )
                { double dist = matches[i].distance;
                  if( dist < min_dist ) min_dist = dist;
                  if( dist > max_dist ) max_dist = dist;
                }

               //std::cout<<"Max dist :"<< max_dist ;
               //std::cout<<"Min dist :"<< min_dist ;

               //-- Draw only "good" matches (i.e. whose distance is less than 2*min_dist,
                //-- or a small arbitary value ( 0.02 ) in the event that min_dist is very
                //-- small)
                //-- PS.- radiusMatch can also be used here.
                std::vector< DMatch > good_matches;

                for( int i = 0; i < descriptors1.rows; i++ )
                { if( matches[i].distance <= max(2*min_dist, 0.02) )
                  { good_matches.push_back( matches[i]); }
                }

                //-- Draw only "good" matches
                Mat img_matches;
                drawMatches( img, key_points, img2, key_points2,
                             good_matches, img_matches, Scalar::all(-1), Scalar::all(-1),
                             vector<char>(), DrawMatchesFlags::NOT_DRAW_SINGLE_POINTS );


              //std::cout<<key_points[1].pt.x<<"\n";
              //std::cout<<key_points2[1].pt.y<<"\n";



   //-- 3. Apply the classifier to the frame


         cv::imshow( "test", img_matches ); //img_matches
        waitKey(30);
 }

return 0;
}

Okay, but how can I get the position ore the place where the most keypoints are. Can someone give me a tip how I must understand it? I saw that I can use something like this: "key_points[1].pt.x" or with y, but do I not must check it to every x,y-place? next would be: good_matches[1].queryIdx but here is it the same question. How can I find where it is?

2. A big question is for me, why is there only a loop over row? Should it not be over row and cols? In my destination it should works like in an array(x,y) and I check every position if it is the same...(have problems with no easy datatypes...)

3. Where I can find/or how the place of the code of drawMatches (for example). Normal way I would try "Open Declaration" (using Eclipse,C++), but I only see the header and not the real function. I need the code and hope I can change all without opencv, or maybe that I can do the loops... and so I must understand how I can read and use the vector DMatch...

Thx for your help. Best Regards,

Answer 1

If I undertand you well you would to find keypoints positions and classifier them. To know the keypoints position you need to do a bucle crossing the keypoints vectors and save the position in a matrix.

Then when you can compare the position of this keypoints with a images areas (whatever you want) and classify them depending on the area in image they are.

Mat pointsInFirstAreaRight, pointInFirstAreaLeft;
    for (int i = 0; i < 2; i++){
           for(vector<DMatch>::const_iterator it = keypoints[i].begin(); it!= keypoints[i].end(); ++it){
                 // Get the position of keypoints
                 float x = [it->queryIdx].pt.x;
                 float y = [it->queryIdx].pt.y;

                 //If the point detected are in the 20 first pixels   
                 if ( (x < 20) && (y > 20) )
                 {
                       //Classify this point in this area
                       if ( i == 0)
                       pointsInFirstAreaRight.push_back(Point2f(x, y));
                       else if (i == 1)
                       pointsInFistAreaLeft.push_back(Point2f(x, y));
                  }
          }
     }

About the sift function, it works doing the detect and extractor at the same time. I use it in my code like this:

>  SIFT sift;
>         
>         /* get keypoints on the images */
>         sift(imagenI, Mat(), keypoints[0], descriptors[0]);
>         sift(imagenD, Mat(), keypoints[1], descriptors[1]);

and then I detect keypoints a extrac the descriptors of points.

Then I only have to do the matching.

I hope to help you.