Inconsistencies and poor facial reconstruction

[ricky4207]

So I am using a 5 camera turntable setup and I am attempting to make full body scans, however the camera alignment is very inconsistent, and on the best ones the face tends to be partially ripped off, or lumpy.  I have attached an example of a great camera alignment and a typical poor camera alignment.  Is there a way to achieve more consistent results regarding camera alignment?  Also what are some tips for fixing the face?  Lastly, how can areas like behind the ear, and behind the hands come out better?  The lighting is fairly even, but since i'm using a turntable some frames may produce shadows behind the ears, under the hands, etc, while other frames don't.  Thank you all!!

(the good camera alignment is the one with the dome -like look, the poor one has cameras scattered all around)

[fabberlounge]

Hi,

Are you masking your shoots, with a sort of a greenbox setup ? Also the position of cameras (stereo pairs, distance about 30cm to 40cm) could help, and a constant light, ( min. two flashes).

Regards, Andreas fabberlounge.com

[Marcel]

This won't help alignment, but I would try and place the top camera slightly lower, so that it would capture more of the face, (should be possible while still capturing enough of the top of the head.)

Currently the top camera is basically rotating in place, and when photos are too close together it can give problems with accuracy on the surface (because the difference in camera position is too small for Photoscan to make a meaningful calculation on the surface).

[admir]

So I am using a 5 camera turntable setup and I am attempting to make full body scans, however the camera alignment is very inconsistent, and on the best ones the face tends to be partially ripped off, or lumpy.  I have attached an example of a great camera alignment and a typical poor camera alignment.  Is there a way to achieve more consistent results regarding camera alignment?  Also what are some tips for fixing the face?  Lastly, how can areas like behind the ear, and behind the hands come out better?  The lighting is fairly even, but since i'm using a turntable some frames may produce shadows behind the ears, under the hands, etc, while other frames don't.  Thank you all!!

(the good camera alignment is the one with the dome -like look, the poor one has cameras scattered all around)

Have you checked sharpnes of images ? Could be head is out of focus. Also clothes worn by subject could be to uniform for photoscan to match points (look point cloud noise in upper body in your first scan compared to jeans).

Admir

[ricky4207]

Hi,

Are you masking your shoots, with a sort of a greenbox setup ? Also the position of cameras (stereo pairs, distance about 30cm to 40cm) could help, and a constant light, ( min. two flashes).

Regards, Andreas fabberlounge.com

Yes I am shooting around 120 images on a white seamless backdrop, and shooting a full round of blank images after to use as a mask using the "from backgrond" option.  I am using 4 flashes, 2 in front both about 45 degrees from the camera rig, same power output on both, large softboxes.  And two in back coming through a large white silk that I am using to blow out the backdrop for easy masking.  I am also aware and careful not to let this light bleed through onto my subject.  The auto masking is doing a great job so far from what I've seen

Have you checked sharpnes of images ? Could be head is out of focus. Also clothes worn by subject could be to uniform for photoscan to match points (look point cloud noise in upper body in your first scan compared to jeans).

Admir

Yes My images are quite sharp.  I'm shooting at f/11, and I could power up my lights easily and go to f/16 or higher, and I'm setting the focus manually before each spin (cameras don't move, subject rotates on turntable).  Now for the clothes being worn I have shot many different styles of clothing, and stuff with more patterns etc. has come out better, but I would like to be able to shoot suits, etc.  and not be completely limited, any way to get around this?

This won't help alignment, but I would try and place the top camera slightly lower, so that it would capture more of the face, (should be possible while still capturing enough of the top of the head.)

Currently the top camera is basically rotating in place, and when photos are too close together it can give problems with accuracy on the surface (because the difference in camera position is too small for Photoscan to make a meaningful calculation on the surface).

I have wondered if the top camera position could be better since it is shooting straight down on the subject.  However in order to do this i would have to back up my turntable further away from my rig, will this be mess up anything with my setup?  I can zoom in much further with each lens to compensate I figure.

Thanks for the responses and help!

Rick

[admir]

I would move top camera to shoot from bottom and move camera that is currently bellow it bit further. From my experience with turntable (4 Canons) there are problematic areas like armpits, crotch, bellow jaw  that more cameras can solve. For top you can use 2 cameras and get good results.

I presume by names of images that you are using Nikons ? If you want more details, i dont think you can get full body scan with 5 cameras.

What we did with 4 cameras is to scan full body (like you did) as reference, then scan separately legs, torso and head and combine those 3 scans using full body scan as reference. It sounds like pain but basically you dont need to move cameras just rotate and zoom.

Here are a few examples:
http://www.agisoft.ru/forum/index.php?topic=1763.msg9342#msg9342
http://www.agisoft.ru/forum/index.php?topic=1763.msg9346#msg9346

[ricky4207]

So I am using 5 canon 5D mark II cameras, and I have moved my turntable back a little bit so my rig doesn't arc over the head like i had it.  I also moved the top 2 cameras back a little bit as well, and have gotten much more improved results.  The face still has some issues.  mostly that the resolution is very low compared to previous attempts.  I have also zoomed in to compensate for moving the turntable, perhaps I need to do facial scans separately like you mentioned admir.

Could you explain the workflow a bit on how to shoot a full body scan as a reference, then combine separate leg, torso, and head scans using the full body as a reference?

attached is a screenshot of the camera positioning now

[admir]

With separate parts you get 3 times more detailed textures and usually 3 times detailed solutions. For our purpose we scanned character in T-pose and posed/animated them later.

What i did (if time permited) was:
For legs parts have chracter wear only pants and boots to get proper deformation over boots and shirt tucked inside pants for easier masking out mesh. Shoot this with legs only visible in images (you can zoom in to get large cover). For this i have rotated cameras vertically to get more coverage.

For torso parts have character wearing what you want and placed over pants as you want to be. Zoom in as much as possible to get full frame of torso part (head doesnt have to be in frames). For this part i have rotated cameras horizontally.

For head i had character wearing gray shirt (something like this: http://wpecdn.com/media/catalog/product/cache/1/image/400x/9df78eab33525d08d6e5fb8d27136e95/i/m/image_64592.jpg which have noise pattern and that helped PS aligning points) and camo beanie like this one: http://www.nexternal.com/armynavy/images/Woodland-Camo-Beanie-Watch-1.gif  if you dont need hair. Those two helped getting alot points and compesate for oilness of skin and hair when aligning cameras and finding points.

For full body pose it as you want wearing same clothes as in separate sessions and align photos. The purpose of this solution to get locations where shoulders, head, waist are and proportions so you can place your separate solutions in same locations.

For full body and separate parts i create final high poly meshes and very low poly mesh (~20 k) for easier manipulation. I use Modo for aligning parts (you can use any 3d app) . You can move loaded meshes in "item" mode around virtual scene. This means actual position of vertices and polygons arent changed only handle so when you replace low poly meshes with high poly you get position, rotation and scale correct.

Sory about crap explanation of workflow , hope it helps.

Admir