PhotoScan Workflow Questions

[cbudin]

Hi,

I am new to PhotoScan, and I have some questions on workflow and how to improve my results.  I've posted some screen grabs below of a simple PhotoScan project I am currently working with.

1) Why did some of the photos/cameras fail to align?  Is there a way to align them manually and/or give the software some hints to help enable alignment?

2) Why did some of the images fail to generate depth maps?  Is this normal or does it represent some kind of problem?  Is there a way to get better results with depth maps?

3) The results on Ultra High still seem a little rough, even in the area with good coverage.  I assume that some of this is due to small movements in the subject while the photos were taken.  Is that the likely cause?  Are there any other causes possible?  Are there ways to improve the results?  Clearly, one way would be to use multiple cameras that capture images simultaneously to eliminate movement of the subject.  Are there other ways?

Thanks, in advance, for any help people can provide.  If you need more information, I am happy to provide it.

P.S. Sorry for the tiny image files, but this forum has very tight restrictions on attachments sizes.  In this day and age 128 KB per image and 192 KB total seems very small!

Clay Budin
NYC

[tezen]

Hello cbudin!
1) In the Pro-Version are markers available. Sometimes just masking (f.e. hairs) will give you good results. The camera-alignment is also very important. Sometimes "low" settings are working better - sometimes not! Take some test-shoots and compare the results by generating them in different chunks with different settings and look for yourself.
2) What camera do you use? For fast model-generation it could be good to disable some photos/depth-maps.
3) Again: What camera do you use? ...and how many Megapixels? 10 Mpix or higher values will give you much better results... take some test-shoots of the same object at different camera-settings, lightning and so on!
(Some people in this forum will disagree but for me cleaning up pictures by denoising-software can give you better results.)
Happy photoscanning... it isn't automatic as it seems to be!
tezen

[cbudin]

I Included an info window for one of the photos in my screen grab (all the others are the same).  The photos are shot with a Canon EOS 60D with 3456x2304 resolution.  I am using the JPG files, not the RAW files from the camera, perhaps that would help?  The ISO is low to minimize noise and the aperture is fairly high to increase depth-of-field.  The shutter is fairly fast to minimize motion blur.

These photos are a sub-set of a larger set which includes the back of the subject.  I am trying to simplify things as much as possible so I can understand the workflow process better.

These photos were taken with a single camera.  The subject turned himself around 360-degrees in roughly 30-degree increments.  Perhaps it should be smaller?  Also there will inevitably be some movement of the subject as he turns.  I was concerned that the subject turning while the background remains fixed might cause problems, but the background is very flat - looking at the matches I see almost no points on the background.  But perhaps masking would help?

Thanks,
Clay Budin
NYC

[meshmaster]

Honestly, ANY movement of the subject will cause issues -- this is the problem with trying to capture humans with a single camera (humans are never, ever still...well, unless they are dead... LOL).  For optimum results you really need a muti-camera setup for any non-static subject.

Kind Regards,

Joe

[Alexey Pasumansky]

Hello Clay,

The movements of subject are really big problem, so in case you are using single camera for face/head capture you need to ask person to remain as still as possible (even breathing and eye movement may cause problems and reconstruction issues). Sitting pose with head fix (for example, head could be leaned on the wall).
Turntable solution for single camera face capture is even more difficult, however, you can refer to experience of other PhotoScan users:
http://www.agisoft.ru/forum/index.php?topic=970.0
http://www.agisoft.ru/forum/index.php?topic=1097.0

Using turntable requires static background masking. If your camera is in a fixed position you can use mask from background feature to automate masking process.

I'd also recommend even greater overlap, 30 degrees is quite close to limit. Another tip is to use hat with good texture pattern, because hair reconstruction and matching is always a problem.

[Heinrich]

I have used a turntable with good results, but I always made sure that the background was uniform (so no points will be found) by placing white cardboard around it, and by making sure that the cardboard is out of focus in the pictures.