Minimal cameras rig for head scanning

[Claude]

Hello! I'm new here.

I would like to make cameras rig for scanning heads. I going to buy Canon 550D with 50 mm lenses.
How many cameras I should buy to make good quality scans of heads.

Thanks in advance for your answer.

[RalfH]

That really depends on what you want to achieve. Do you want to capture the full head or only the face? What is "good quality" for you? I think that spacing the cameras at angles of approx. 15° would be a more or less ideal scenario, which means you'd need 24 cameras for a full circle. In addition, you will need some looking down from above and some looking up from below (6-12 each). Of course, 36-48 cameras are a lot of money, but this is the ideal scenario. You might get by with leaving out every second camera (i.e. a total of 18-24 cameras), but I don't think that you'll be very happy with less. Also, of course you'd need less if you only want to capture faces. And, again depending on what you want to achive, you might want to consider whether a DSLR is truly necessary or whether a number of point-and-shoots would suffice.

[Claude]

Thanks RalfH for answer!

I want to get best quality as it possible. The most important is face but it will be nice to have also scanned back of had even with less precision.

I made quick symulatin in photoshop and made half of camers with distans 15° and rest 30°. I need for it 18 cameras.

 Now I have to find out how many cameras I shuld put up and below head.
I put 5 cameras below and 5 cameras over head.

Finnally I need 28 cameras. What do you  think about such setup?

[RalfH]

I think the setup looks ok except for the upper part of the head and the back of the head where you should probably add another two cameras to cover those areas.

[andy_s]

Well for me - it's whole body scanning but a full DSLR rig breaks the budget.

So, thinking of head/shoulders with multi-dslr [maybe 24MPix Nikon D3200] maybe hung off a geodesic dome and point/shoot for body/legs. Approx 100 microsecond capture using studio flash is planned.

Two related thoughts for which i'd appreciate any comment:

(i) since the sections will have different pixel sizes might photoscan have "problems" in it's epipolar search algorithms [if that's the way it works  ] in the overlapping different pixel size frames ?

(ii) anyone comment on satisfactory / otherwise results obtained from e.g:

Canon G10[14.7MP], Canon A3000IS [10MP], Canon SX110IS [9MP], other ??

I am mindful of the comment at:

http://www.agisoft.ru/forum/index.php?topic=942.msg4638#msg4638 

as follows:

Infinite Re: Samsung NX1000
« Reply #5 on: January 20, 2013, 01:42:51 PM »
I can honestly say from experience dating back to 2008 (stereo photogrammetry using pairwise cameras with Dimensional Imaging software, before using Scanner Killer by xyz-rgb) that anything below 18MP really isn't great. You get OK results with 12MP but it's not ideal...Prime lenses are best for image quality. The key is sharp, focused, contrasted images. As much detail as possible, for good reconstruction.

[andy_s]

I made quick symulatin in photoshop and made half of camers with distans 15° and rest 30°.

Hi Claude,

I also (vaguely) remember reading that photoscan likes pairs of images and "filler" shots - so equi angles may not be optimal...

[Claude]

I have to fit to budget. I reduce rig to 12 cameras.

andy_s - Yes I notice that people use two cameras for one part of scanning object.

I have limited budget so I wonder if with 12 cameras it is better to put it in pairs and scan half of head or use like on my simulation around head.

My simulation. I will render soon and test result in Agisoft

[andy_s]

I wonder if those "pairs" are 100% overlap and placed at the more complex surfaces:

http://www.int-arch-photogramm-remote-sens-spatial-inf-sci.net/XL-5-W1/251/2013/isprsarchives-XL-5-W1-251-2013.pdf

"2.2 Large baselines and surface tilts. In order to evaluate dense surface reconstruction for large baselines and intersection angles (e.g 90 degrees), the matching performance is decreased for large intersection angles since the image similarity is reduced. This image similarity is mainly dependendent on this angle , but also the angle between normal vector and viewing ray. The latter is particularly important for complex surfaces, where surfaces are often tilted more than 20 degrees in relation to the viewing direction and no observation of these surfaces in nadir direction is possible. Thus, flat surfaces can be matched successfully on larger angles than tilted surfaces."

[James]

Just a thought, i have never used a multi camera setup or much in the way of head scanning, but:

Assuming you are scanning human heads, the only part of the head subject to movement is the face so set up your rig to capture the face only, then you can move the whole rig to capture the sides, back and top, which are unlikely to have moved.

Obviously the head itself will be moving all over the place, but so long as the background is either plain or masked out then photoscan would never know.

[Infinite]

I doubt you will be able to get very good results with 12 cameras trying to do 360. You would certainly need structured light in there from multiple angles to help reconstruction.

[Claude]

Hello Infinite!

So it will be better make 6 stereo pair and scan only face than whole head with worse quality?
Wow big should be the angular distance between the camera?

Structured light - you mean projection noise on model? Textures are less important for me, so it could be good solution.

Thanks in advance for your answer.

I doubt you will be able to get very good results with 12 cameras trying to do 360. You would certainly need structured light in there from multiple angles to help reconstruction.

[Infinite]

Hello Infinite!

So it will be better make 6 stereo pair and scan only face than whole head with worse quality?
Wow big should be the angular distance between the camera?

Structured light - you mean projection noise on model? Textures are less important for me, so it could be good solution.

Thanks in advance for your answer.

I doubt you will be able to get very good results with 12 cameras trying to do 360. You would certainly need structured light in there from multiple angles to help reconstruction.

I would say experiment. Try it digitally first as you have with projected noise, see what results you can get. You will also need to test for real. Try and master 180 first in the real world, swap cameras positions. See what works best for you. There is no manual for this, each person has their own way of doing things