Trying to find chekerboard points, not returning any points with detectCheckerboardPoints

Hi all

I have a fisheye lens system through which I'm taking images. The images are highly distorted, but there are also many of these images, and I wanted to find an automated way of consistently undistorting them. I printed out a checkerboard pattern, and took multiple images of this pattern, a representative image shown.

However, when I try to use the detectCheckerboardPoints, I get no points back. I know that this algo is sensitive to noise, which I definitely have in my images, but I can't do a whole lot better with my current system, so this is what I've got. I've tried playing with HighDistortion toggle, the MinCornerMetric, which I set as low as 0.0002 and as high as 0.5, and the PartialDetections toggle, to no joy. Can anyone suggest further steps, or alternative algorithms, or a better way of doing this? Otherwise, I'll have to parse through hundreds of images and manually label the points, and that is likely to take a while.

Some of the pre-processing steps I've tried:

1. imbinarize
2. imerode
3. medfilt2
4. Combination of the three above steps in varied order

Answer 1

[Matt J]

You can download pgonCorners,

https://www.mathworks.com/matlabcentral/fileexchange/74181-find-vertices-in-image-of-convex-polygon

However, it will only find the corners of one chequerboard square at a time, so you will have to extract separate images of each black square:

load chequerboard

BW=~imfill(BW,[1,1]);
BW=bwareaopen(BW,10);
BW=imclose(BW, ones(5));
BW=imerode(BW,ones(3));

rc=fliplr(vertcat(regionprops(BW,'Centroid').Centroid));

clear V
for i=height(rc):-1:1

 onesquare=bwselect(BW,rc(i,1),rc(i,2));

 % imdisp(onesquare)

 V{i}=fliplr(pgonCorners(onesquare,4));

end

V=vertcat(V{:});

imshow(BW,[]);hold on
scatter(V(:,1),V(:,2),'r','filled','SizeData',15); hold off

[Alexander Dumont]

Hi Matt

Wow that seems to work very well for you. I tried to copy over your script, and I am still getting nothing, this time giving an error at V{i}=fliplr(pgonCorners(onesquare,4));

%%edit

Actually, I am getting several but not all vertices I need. It looks like this:

Could you tell me a bit more about how you pre-processed. Here is my script as it is now. It works as expected when I load checquerboard, but not when I try to use the original image.

If not, I think I can get somewhere with everything you've provided, it's a very good start point. Thank you for all the help!

 clear
clc
close all

d = dir('calibrationImages\*.png');

for j = 1:length(d)

    figure
    subplot(121)
    I = imread(d(j).folder + "\\" + d(j).name);
    imshow(I)
    title('original image')

    %I = imadjust(I);
    BW = imbinarize(I);
    n = 500;
    BW = imresize(BW,[n n]);

    %load chequerboard.mat
    BW=~imfill(BW,[1,1]);
    BW=bwareaopen(BW,10);
    BW=imclose(BW, ones(5));
    BW=imerode(BW,ones(3));
    
    subplot(122)
    imshow(BW)

    rc=fliplr(vertcat(regionprops(BW,'Centroid').Centroid));
    clear V
    for i=height(rc):-1:1
        try
            onesquare=bwselect(BW,rc(i,1),rc(i,2));
            %imdisp(onesquare)
            V{i}=fliplr(pgonCorners(onesquare,4));
        catch
        end
    end
    V=vertcat(V{:});
    imshow(BW,[]);hold on
    scatter(V(:,1),V(:,2),'r','filled','SizeData',15); hold off

end

[Matt J]

You need to attach the image that fails (post-binarization).

[Alexander Dumont]

Here is the image post-binarization

[Matt J]

Since you accept-clicked the answer, I assume you overcame the problem?

[Alexander Dumont]

I think I'm at a point where I can figure out the rest. Thank you for all the help, it saved me days of hassle.