No success with alignment of 12mm lens images

[jupp2000]

Buongiorno,
i have the 15 day trial here (again), but i dont get any successful results in photo alignment at all.
In use is a 12mm Samyang lens (no fisheye). That lens gives wonderful results with my photosets in another photogrammetry software, but with zephyr it is not possible to work with.
I always photograph the front of buildings, shop windows etc. so not completely around the buildings.
It works great with the other trial software, but i like the zephyr interface and would actually like to buy the software.
Are there any tipps?

Kind regards
jupp2000

[Andrea Alessi]

Hi jupp,

it's possible we have a wrong pre/calibration in our database, if it straight doensn't work with zephyr but it does in other software. It's not very common, but it can happen.

May I ask you to share your dataset with us at support@3dflow.net so we can take a look?

Alternatively, you can try running the dataset by unchecking the "use online calibration" option in the wizard (it is on by default) so you will use the fully full autocalibrated functionality. However, we'd appreicate the photos if possible so we can see and prevent problems to other users that may be using your same device.

[jupp2000]

Thank you Andrea, for your fast responding,
I am not shure, what you mean with dataset.
Do you want the images? No problem, with google drive.
Or any other infos or statistics from inside Zephyr?
In the meantime i had a "little" success with calibrating the lens, but there are misalignments in the point cloud anyway.



Thanks in advance!
Jupp2000

[Andrea Alessi]

Yes, images and eventaully external data you may have made/used (masks, zep project) but tat th every minimum, I require all the photos in the project

A google drive link works, thank you!

[jupp2000]

Email is on the way!

[Andrea Alessi]

Hi jupp,

here is my advice to improve your dataset:this is one of the cases where you may get decent results with a software, and not in another, as the photos are far from ideal unfortunately. You may find that the software package X works better than Y in this specific case due to the quality of the images, and in other limit cases the other way around.

Another solution would be to mask out the background from your photos, but this still would leave you with a very low amount of information to work with. I would recommend anyways to simply reshoot this subject.

So, here is a quick and dirty "proof of concept" on how to work with this dataset, although I strongly recommend reshooting this in a more proper way.

I did a very quick masking, since you more or less kept the subject in place. I used the rectangle tool and uset the copy to all (so masking took all about 30 seconds)



As you can see the orientation seems correct now, although there is still a lot of noise coming in from the background.

This translates to still some orientation issue, if you go and take a look with control points (full version only) you can immediatly see there is still a 2-3 pixel reprojection error, which is quite a lot.

You could at this point add 2D control points and run the bundle adjustment to fix the orientation issue, as otherwise you'll get a very noisy mesh.

So in short, if you want this dataset to work:


- make sure to mask out your images. The better the mask, the better the results - use the blue/red strokes tools to create quickly masks that can help you remove the background. This could probably take about 10-30 seconds per photo, rather than my 30 seconds total, but will yield better results
- consider doing a manual calibration of your lense, since this dataset is problematic (make sure to use the very same settings, focal lenght and resolution)
- if you have the full version of zephyr, add control points and use them as constraints for a bundle adjustment (tool - workspace - optimize camera parameters)

Although I ultimately suggest you take your photos by making sure the images are crisps and clear, with no iso speed noise, and that your subject is framed for as much as you can in your photos.

I hope this helps!

[jupp2000]

[jupp2000]

I am back and can tell, that it is working much better now.
Like you advised, masking does the trick. I hate masking, but it seems to be unavoidable.
The 12mm lens seems not suitable for narrow roundup shots like my example above.
I shot another set (Fuji 18-35mm @ 18) with a turntable and masking afterwards and the alignment and modeling is working!
Thanks again Andrea!