Exception during sub-marker detection

[abrock]

The demo tool now works fine with the images from #1 but dies with an exception when fed my new images, so I made a new set of cropped images showing the problem:

input_Cam011-sub2.txt

input_Cam011-sub6.txt

test-sub-mitte.txt

test-sub-mitte-mitte.txt

test-sub-mitte-mitte-unten.txt

test-sub-mitte-mitte-unten4.txt

test-sub-scaled.txt

test-sub-scaled-links.txt

test-sub-scaled-rechts.txt

[abrock]

I wrote a tool for finding the smallest image reproducing the problem:

1. run extractMarker on the input image.
2. if an error occurs subdivide image in 3 overlapping images [0-1/2], [1/4-3/4], [1/2-1]. The axis of the subdivisions is the longer axis of the image.
   2.a) run the tool on all three of the sub-images.

webcam.zip

You can find the tool here: https://github.com/abrock/hdcalib (localize-exception.cpp)

I'm running the tool on images with higher resolution (~30MP) but that will take a while.

[abrock]

High resolution process still running, here are the relevant results from a downsampled (50%) version.
Github wouldn't let me upload the complete dataset but I think the samples will be sufficient.
Tell me if you need all of them.

test50.png-l-m-m-m-b-m.txt

test50.png-l-m-m-m-m-m-m.txt

test50.png-l-r-m-b-m-m-b.txt

test50.png-l-r-m-m-b-m-m.txt

test50.png-l-r-m-m-m-m-b-m.txt

test50.png-m-l-b-m-m-m-m-m-m.txt

test50.png-m-l-m-m-b-m-m-b-m.txt

test50.png-m-l-m-m-b-m-m-m-m.txt

test50.png-m-l-m-m-m-m-m-m.txt

test50.png-m-l-m-m-m-m-m-m-b.txt

test50.png-m-l-m-r-m-m-b-m.txt

test50.png-m-m-l-b-m-b-m.txt

test50.png-m-m-l-m-m-m-b-m.txt

test50.png-m-m-m-m-l-m-m-m.txt

test50.png-m-m-m-m-l-t-m-m-m.txt

test50.png-m-m-m-m-m-m-l-m.txt

test50.png-m-m-m-m-m-m-m-t.txt

test50.png-m-r-m-l-m-m-t.txt

[abrock]

The complete results for the 25% downsampled input image:

25.zip

[abrock]

I have the results for the original resolution.

In the off_hdm file: Is the green cross the final position or the initial position? Some of those are way off any dot on the target.

Those are the smallest results which reproduce the error:

And finally the original image: