[mexopencv] Epipolar Geometry example

README.md

Epipolar Geometry example

MATLAB code to show epipolar geometry with OpenCV (using mexopencv), based on code from OpenCV-Python Tutorials.

example_epipolar_geometry.m

function example_epipolar_geometry
    % a pair of stereo images (grayscale)
    img1 = cv.imread('left.jpg', 'Flags',0);
    img2 = cv.imread('right.jpg', 'Flags',0);

    % detect keypoints and calculate descriptors using SIFT
    [key1,feat1] = cv.SIFT(img1);
    [key2,feat2] = cv.SIFT(img2);

    % match descriptors using FLANN
    matcher = cv.DescriptorMatcher('FlannBased', ...
        'Index',{'KDTree', 'Trees',5}, 'Search',{'Checks',50});
    m = matcher.knnMatch(feat1, feat2, 2);
    m = cat(1, m{:});    % N-by-2 array of structs

    % keep only "good" matches (using ratio test as per Lowe's paper)
    idx_good = ([m(:,1).distance] < 0.8*[m(:,2).distance]);
    m = m(idx_good,1);

    % extract keypoints from filtered matches
    % (0-based vs. 1-based indexing in C++/MATLAB)
    pts1 = num2cell(cat(1, key1([m.queryIdx]+1).pt), 2);
    pts2 = num2cell(cat(1, key2([m.trainIdx]+1).pt), 2);

    % find Fundamental matrix
    [F,mask] = cv.findFundamentalMat(pts1, pts2, 'Method','LMedS');
    mask = logical(mask);

    % select only inlier points
    pts1 = pts1(mask);
    pts2 = pts2(mask);

    % find epilines, and draw them in corresponding images
    lines1 = cv.computeCorrespondEpilines(pts2, F, 'WhichImage',2);
    lines2 = cv.computeCorrespondEpilines(pts1, F, 'WhichImage',1);
    [img11,~] = drawlines(img1, img2, lines1, pts1, pts2);
    [img22,~] = drawlines(img2, img1, lines2, pts2, pts1);
    
    % result
    figure('Position',[200 200 1200 400])
    subplot(121), imshow(img11)
    subplot(122), imshow(img22)
end

function [img1,img2] = drawlines(img1, img2, lines, pts1, pts2)
    c = size(img1,2);
    img1 = cv.cvtColor(img1, 'GRAY2RGB');
    img2 = cv.cvtColor(img2, 'GRAY2RGB');
    for i=1:numel(lines)
        clr = randi([0 255], [1 3], 'uint8');
        r = lines{i};
        p1 = uint32([0, -r(3)/r(2)]);
        p2 = uint32([c, -(r(3)+r(1)*c)/r(2)]);
        img1 = cv.line(img1, p1, p2, 'Color',clr, 'LineType','AA');
        img1 = cv.circle(img1, pts1{i}, 5, 'Color',clr, 'Thickness',-1);
        img2 = cv.circle(img2, pts2{i}, 5, 'Color',clr, 'Thickness',-1);
    end
end

Thanks for the code

Hi,amr Can you help me with that issue ,http://stackoverflow.com/questions/28320331/how-can-i-design-an-accurate-mosaic-from-homography I have opened this problem in stackoverflow

I have a problem about that codes.I've error using imread failed.Could you help me please?

Error using imread
imread failed

Error in example_epipolar_geometry (line 3)
img1 =cv.imread('left.tif', 'Flags',0);