We are running a rotary scanner which has 7 rows of cameras. When we run the alignment process, the most top row cameras are aligned together in a single component and not aligned with others, hence do not taken into account for the further steps (mesh, texture generations), which results to a bad quality head part (see pictures).
I am sure that those cameras have features connected to lower level ones, but none of them detected. Why does this happen? Is there any setting that we can consider to change to get a better result?
P.S. I’ve uploaded the photos and aligned xmp files to the box repository that you’ve provided inside Top cameras not aligned folder, for you to have a look.
Thanks.
Hi Serob,
as it is still the same rotary scanner, I suppose there will be still some issues with that.
I was able to align the data in two steps.
Firstly, I imported groups 6 and 5 and aligned. It was successful. Then I imported all remaining images and aligned again, which was successful.
Why did you chose this approach, what’s the logic behind it? Wouldn’t there be one step possibility of having them all (or most of them) aligned? 'Cause that’s very strange when only 6 and 5 can be aligned, but when there are all groups, one of them can not…
I tried different approaches and then I aligned only the 6 line and it was quite deformed…
So I tried to add the closest one and it was possible to align without any issue.
It is true that it should be aligned using the whole dataset, but as you know, this is not an ideal capturing way, so there could be some errors.
And it looks like there is something wrong with line 6, as using only those images it provides deformed alignment. Is there used something different to the other cameras?
Also, like the data 1, 2, 3, 4, 56 works to align (but not always), The only problem there is combination of 0 and 6 images.
The problem could be related to the insufficient number of features over the shirt area:
What is deformed there on the 6-th set? I mean how did you notice that it may be different from others? It is the same camera model with the same settings as others, so don’t know what can be different here.
And if there’re insufficient number of features over the shirt area, why the other cameras pointing to shirt can align better?
I’ve succeeded to aligned all images in one step by increasing Max feature reprojection error from 2 to 5, and/or by changing Image downscale factor from 1 to 2.
So the questions are:
What drawback can increasing Max feature reprojection error have, as it is written in the documentations that recomended value is 3 mpx
Why increasing Image downscale factor improves the aligned images number (this is true for all the type of the scanners and cameras we used and continue to use)? Logically thinking - as less downscaled images you have as higher details there are, hence the alignment should be better, but that’s not the case. Why?
Not sure why, but in this case the image is like more blurred, so maybe there are found some features considered as the same and therefore the images are aligned.
The result that all images are aligned doesn’t mean that it is a successful alignment. There could be some errors in the camera poses, as you used “weaker” settings.
