NikolasE's profile - activity

"Which of these values are important for the BFMatcher or the FlannBasedMatcher to compare the images. " Non of these values, in fact. You only show the adresses (pointers) of your data and not the descriptors. You only have to save the descriptor matrix (and the keypoints).

I havn't tried it myself, but there is SphericalWarper in 3.0.0 http://docs.opencv.org/trunk/modules/stitching/doc/high_level.html#sphericalwarper Have you tried it?

Can you show some sample images?

The c920 is a really good camera, setting focus is possible. (In Ubuntu, I use v4l2_control and it works reliable)

How does your thresholded image look like? Have you tried to simply downsample it?

Could you post some example images? To get the red areas, you could do something like described here: http://stackoverflow.com/questions/10948589/choosing-correct-hsv-values-for-opencv-thresholding-with-inranges

After that, you detect connected components, discard small components and the fit an ellipse to the boundary pixels. (all described here http://docs.opencv.org/modules/imgproc/doc/structural_analysis_and_shape_descriptors.html )

BTW:

Do the markers have to be red? I'd guess that e.g. green dots would be easier to detect on human skin.

What have you tried to write and read the image? I work with a carmine, which also uses 16UC1 for depth and I write with cv::imwrite and read via

cv::Mat depth = cv::imread(filename,CV_LOAD_IMAGE_ANYDEPTH | CV_LOAD_IMAGE_ANYCOLOR);
assert(depth.type() == CV_16UC1);

This is basically an image stitching task so you could start here: http://docs.opencv.org/modules/stitching/doc/introduction.html You could use Sift/Surf/etc to find corresponding points and then compute the homography between the two images. You can then get the intersection polygon by applying the homography to the corners of one image.

I have had the same problem and solved it by upgrading to the dev-version of opencv (2.9).

I have the same problem using Ubuntu 12.04 and OpenCV 2.4.6. Using guvcview, I can set the image size to up to 1920 x 1080 but I can only read VGA images in OpenCV.

The camera is opened as: cap = VideoCapture(dev_id); assert(cap.isOpened());

// returns 1
std::cout<< cap.set(CV_CAP_PROP_FOURCC,CV_FOURCC('M','J','P','G')) << std::endl;

// returns 0
std::cout<< cap.set(CV_CAP_PROP_FRAME_WIDTH, 1280)         << std::endl;
std::cout<< cap.set(CV_CAP_PROP_FRAME_HEIGHT, 720)       << std::endl;

Has anyone managed to get larger images in OpenCV?

Hey!

I just upgraded to 2.4.5 and like the new features of the highgui as saving the image or zooming. But I also have some problems:

The first problem are missing icons in the window's toolbar. The buttons exist are work, but they have no icon so that they are basically invisible.

The second problem is visible if I zoom far into a very small part of the image. At some point, each pixel is surrounded by a black border and contains three printed values: a red 0, a green 48 and a blue 57. I guess that these values should correspond to the color of the pixel (which is by the way correctly printed in the window's statusbar).

Here is my OpenCV cmake-result:

-- GUI: -- QT 4.x: YES (ver 4.8.1 EDITION = OpenSource) -- QT OpenGL support: YES (/usr/lib/x86_64-linux-gnu/libQtOpenGL.so) -- OpenGL support: YES (/usr/lib/x86_64-linux-gnu/libGLU.so /usr/lib/x86_64-linux-gnu/libGL.so /usr/lib/x86_64-linux-gnu/libSM.so /usr/lib/x86_64-linux-gnu/libICE.so /usr/lib/x86_64-linux-gnu/libX11.so /usr/lib/x86_64-linux-gnu/libXext.so)

(I also used QT 4.8.1 to build my project. I used QtCreator on a Ubuntu 12.04 machine and g++ 4.6.3)

Third parties only Eigen and TBB.

Do the icons maybe only exist in a specific QT-Version?

Cheers, Nikolas

Update: The Icons are no QT-Icons but come with openCV. The are located in modules/highgui/src/files_Qt/Milky

and opencv-2.4.5/modules/highgui/src/window_QT.qrc contains mappings from zoom_x1-icon to a file in this folder (in which the image names are just numbers as 42.png).

no, same behavior. (I just get a VGA image)

Hello

I just reinstalled OpenCV to activate TBB and also added the unicap option during cmake. So far, I used a webcam that can provide 1280x720 images. So far, I used

  cap = VideoCapture(1);
  cap.set(CV_CAP_PROP_FRAME_HEIGHT,720);
  cap.set(CV_CAP_PROP_FRAME_WIDTH,1280);

to connect to the camera which worked perfectly. If I now try to open the stream as before, no Image is read (cap.read(rgb) does not return). If I uncomment the cap.set lines, I get a 720p image, but only the upper left 640*480 pixels contain the full image and the rest is just black.

I am not sure if unicap has something to do with it, so it's just a guess. Has anyone experienced this problem before?

Cheers, Nikolas

cv::flann stores the index as a data-file so that diff can only tell that both files are different: Binary files flann_copy.txt and flann.txt differ

Hello

I am trying to store and load flann::indices to speed up some experiments. I extract features and descriptors using OpenCV 2.4 and SurfFeatureDetector/SurfDescriptorExtractor, create an Index and save everything:

    Mat img = imread(img_file.toStdString(),CV_LOAD_IMAGE_GRAYSCALE);

    SurfFeatureDetector detector(minHessian);
    SurfDescriptorExtractor extractor;


    vector<KeyPoint> keypoints;
    detector.detect( img, keypoints);

    Mat descriptors;
    extractor.compute( img, keypoints, descriptors);

    flann::Index ndx(descriptors, flann::KDTreeIndexParams(4));

    QString kpts_file = sub_folder + "/keypoints_descriptors.yml";

    FileStorage fs(kpts_file.toStdString().c_str(), FileStorage::WRITE);

    cv::write(fs,"descriptors",descriptors);
    cv::write(fs,"keypoints",keypoints);

    fs.release();

    QString ndx_file = sub_folder + "/flann.txt";

    ndx.save(ndx_file.toStdString().c_str());

So far, everything works fine. To check if everything is okay, I read the files again and write the index again into a file:

FileStorage fs(kpts_file.toStdString().c_str(), FileStorage::READ);

fs["descriptors"] >> info.descriptors;
cv::read(fs["keypoints"],info.keypoints);

QString kpts_file2 = sub_folder + "/keypoints_descriptors_copy.yml";

FileStorage fs2(kpts_file2.toStdString().c_str(), FileStorage::WRITE);
cv::write(fs2,"descriptors",info.descriptors);
cv::write(fs2,"keypoints",info.keypoints);

info.flann_tree.load(info.descriptors,ndx_file.toStdString().c_str());

QString ndx_file2 = sub_folder + "/flann_copy.txt";


printf("Img: %s, kpts: %zu\n", img_name.toStdString().c_str(), info.keypoints.size());

info.flann_tree.save(ndx_file2.toStdString().c_str());

I then compute the md5sum of the two files I wrote. The yml-File containing the keypoints and the descriptors have the same checksum, but the two flann-files don't match (although they have the same size).

Can someone tell me what happens during saving and loading that could change the flann::Index?