Smart Shooter, Nikon D3200 and noiseprojection.

[Ecsplorer]

Hi Magnus, hi all,

pretty interesting thread.

I wonder how do you sync the projectors and how do you control delays and such?
Would you mind to share?

Best regards

[Magnus]

Hello Ecsplorer!

We don't really do any sync with the projectors, just split the signal from one computer.
Delays will be present in the noisecapture due to the 1100Ds having somewhat inconsistent shutter lag (time from pressing trigger to shutter activating).
For example when we tried syncing with flash and having the shutter at 1/50, 2 or 3 cameras (out of 12 or 14 1100Ds I think it was) had black images each time but it was random which cameras lagged, hehe.
3 more D3200s arrived today and we will try some more tests in the coming days.
And going forward the way we will control delays is using something like Camera Axe http://www.cameraaxe.com/ (possibly their Multi-Flash Plus too).
We are thinking of using the light trigger and connect it to the Canon group so that the Nikons trigger together with flash and then the flash activates the trigger for the Canons. What we will see is if the shutter lag is short (it will most likely be over 40ms though) enough to not induce too much movement error.
Another way would just be to have the Canons set to a delay that insures that they will trigger as shortly after flash as possible.

Best, Magnus.

[Ecsplorer]

Hey Magnus,

thanks! The Camera Axe things seems really interesting.

[andy_s]

Hi Escplorer,

it does appear to be very good - and very expandable. You can even measure shutter lag without the need for an oscilloscope:

http://www.techphotoblog.com/tpb-13/

[tmpdev]

@Magnus: We will process at High setting since the Ultra High takes too long and we retrieve the small details from the texture map as can be seen in this one.
https://www.dropbox.com/s/uj872lln5lth30f/130615_ZBtest_01.jpg
For this I decimated the model down to 200000 in PS and then I just did smoothing on it and no other cleanup. I am a beginner when it comes to post-processing so keep that in mind

Hello Magnus, This is exactly what I am interested in achieving using Nikon as well. I am very pleased with your end result in Zbrush! I know you said you used smoothing only(forgive any question that may seem dumb)...was that for the imported obj mesh? How many cameras did you use to get the head geometry you imported into Zbrush(the screenshot)? I am new to Zbrush, but it will obviously be a critical part of my workflow. With regard to Smart Shooter, does it fire multiple cameras at once? Do you hook all usb connections to a usb repeater and then to the computer? Please explain your setup in detail in it's not problem. And lastly, I think I understood you to say that for facial capture, there is no need for a projector.

Kind Regards,

tmpdev

[Magnus]

Hello tmpdev!

For that one it was mostly Canons actually, hehe. Here you can see the layout of the cameras.

https://www.dropbox.com/s/dkd7rk8xlqje22y/130613_Magnus_Test_01.jpg

A you can see this does not have total coverage but we were just doing some rough tests during this shoot (you will for example see some noise below the chin in the next pictures). This is not the best layout either, the results were a bit noisy due to distance between cameras and zoom. It does give a hint of what can be done though.
Having stereopairs close together for the face in particular is something we've found to give better results. Then depending on how much you need of the hair/back of the head you could increase the distance between the cameras in the back.

Here is the mesh decimated down to 200000 in PS.

https://www.dropbox.com/s/28obvd01awmjfbe/130613_Magnus_Test_02.jpg

!Word of caution: I am basically just messing about in ZBrush and I am a novice with it so this is most likely a non-optimal workflow, hehe. Maybe someone else can chime in on some pointers.!

Then I imported into ZBrush, did "Polish Crisp Edges", removed some noise from the chin with Smoothbrush, DynaMesh and UV Master. Next I re-imported into PS and it looks like this (smoothing a bit too harsh maybe but this was just a test ).

https://www.dropbox.com/s/3u4k2tkd29l6mg8/130613_Magnus_Test_03.jpg

I then just made a texture with Keep UV option and imported it into ZBrush.

This is how the model looks in ZBrush.

https://www.dropbox.com/s/u4dzoxrg3yi9yqy/130615_ZBtest_02.jpg

Then I Subdivide the mesh, make a mask from the texture and then I go into Deformation and Size and get this result. On another test not shown here I did some tricks in Gimp and I turned the facialhairs into white to get them to go outwards.

https://www.dropbox.com/s/p92r44wr8uycatc/130615_ZBtest_03.jpg

Again I am just showing the way I am playing around in ZBrush, hopefully someone with more knowledge can add some better information.

The triggering of the cameras we do with a simple corded shutter release and a bunch of 3.5mm audio splitters, Smart Shooter we use to change settings and get the pictures off the cameras into the computer. We have the cameras connected to unpowered USB hubs and then to a powered usb extender cable going into a laptop.
For this particular shoot we used a Speedlite on one of the Canons set to 1/20th shutter with second curtain shutter and the rest of the cameras at 1/5th shutter (so that the 1100Ds can trigger within the timeframe of the flash). The Speedlite in turn triggers the other three studioflashes we use (one laying down behind me with showercurtain over it, hehe). We shoot two of the flashes thru white cloth (IKEA showercurtain ) to get diffusion. This is due to current lack of space, we will have more flashes with diffusion around the subject.

The harsh light of the projectors is something that makes it less suitable (in my limited testing) for capturing the face, it does not really add anything if you can get sufficient detail without it.

We have some plans to try infrared and maybe capturing UV... Hopefully we have something we can show soon.

Best, Magnus.

[scanlab.ca]

thanks for sharing your info Magnus!
hi just a quick question, why not trigger with smart shooter instead of remote?  i'd like to be able to software trigger canon and nikon dslrs.

[Magnus]

Hello edgeArchitect!

The problem with triggering over USB is that it is one camera after the other, so quite a lot of delay introduced (the technical details have been discussed in this forum before, can't remember where though, hehe). As far as I know there is no way to trigger thru software without significant delays.

Best, Magnus.

[scanlab.ca]

thanks Magnus! yes we are having that problem with trigger delay. appreciate the info!

[jeancampos]

Hello Magnus

I am very grateful for the help you have given to beginners like us.
I want to share with you the great results with our Rig 13 X Canon 1100D using noiseprojection with a projector EPSON 800x600. We had a fantastic difference of the results of the cloud in PS.


Result without the projector:

https://dl.dropboxusercontent.com/u/10237587/Artes_Web/Forum/Agisoft/Screenshot%20on%207.1.2013%20at%2010.17.26%20AM.png

Results using the projector:

https://dl.dropboxusercontent.com/u/10237587/Artes_Web/Forum/Agisoft/Screenshot%20on%207.1.2013%20at%2010.17.15%20AM.png

We are buying a HD projector, and the used here was borrowed rs. I think the next test will be more complete. Our rig allows us to only 180 ° for facial scan, we are buying 19 more DSLRs to be able to complete the 360 ° and thus get a more complete result. Magnus, I researched thoroughly the Nikon D3200, and I believe we will make a Hybrid Rig, our 13 Canons + 19 Nikons, I expect a more refined result, for pores and small details of expression, something that Canon still with no success.

I'm also thinking about how to solve the problem of texture. In this section we had to process the 24 images, and 12 images with projector to generate the model, and 12 images without it on only to generate the texture, unfortunately the PS does not allow designing the texture images added later output would be the same for the projection Zbrush and Mudbox. I saw that Lee suffered the same problem, and I wonder how to solve.

Here, the refered pictures.

https://dl.dropboxusercontent.com/u/10237587/Artes_Web/Forum/Agisoft/Screenshot%20on%207.1.2013%20at%2010.18.12%20AM.png

https://dl.dropboxusercontent.com/u/10237587/Artes_Web/Forum/Agisoft/Screenshot%20on%207.1.2013%20at%2010.19.21%20AM.png

Testing continues. I'm thinking of testing small objects and see the possible results.

Thanks to all!

[Magnus]

Hello Jean!

Great tests! It can make quite a difference.

The Canons should be good for human faces though (without any projection).

Do try different distances between the cameras, it can make quite a difference.
As you can see in this test with 6 1100Ds that by keeping them close together (kind of like stereopairs) you get a lot of detail.

https://www.dropbox.com/s/w9vy0u9jtn3opfv/130322_Magnus_Test_01.jpg

This is how the cameras were set up.

https://www.dropbox.com/s/y7ek6z2hn9f96t5/130322_Magnus_Test_02.jpg

Also mixing different crops can work pretty well.

https://www.dropbox.com/s/0k4ofd3zeh24cvb/130419_Magnus_Test_01.jpg

https://www.dropbox.com/s/fhqh312ztswpy7m/130419_Magnus_Test_02.jpg

I will hopefully this week get the rig set up, we've taken it down to change the mounting of the cameras (will show soon, it was the cheapest way I could find Manfrotto Super Clamp, some steel plates and some screws, hehe).

I have a lot more things coming up, hehe.

Something I am thinking of trying is converting camera(s) to full-spectrum. That way I can shoot IR with an IR filter and then still use it as a normal camera with a color correcting filter.
See this video as an example.

http://www.youtube.com/watch?v=RpYRraBniOA

Then I will use several Kinects (or Xtions) as IR noise-pattern projectors. Here you can see what its pattern looks like (scroll down quite a bit).
 
http://www.futurepicture.org/?p=97

This I think could be quite an interesting test, might even try some 4D with that, hehe.

Best, Magnus.

[Infinite]

Then I will use several Kinects (or Xtions) as IR noise-pattern projectors. Here you can see what its pattern looks like (scroll down quite a bit).
 
http://www.futurepicture.org/?p=97

This I think could be quite an interesting test, might even try some 4D with that, hehe.

Not meaning to burst your bubble Magnus but it might save you some money. Here are some 4D examples I took back in early 2011 using 2x stereo pairs.



Sadly the Kinect IR pattern really isn't suitable for scanning, more so for motion capture. The laser is diffracted through a hard etched patten and prime sense aren't intereted in modifying the pattern unless someone can place a multi-million unit order : / There are IR DLP projectors on the market but they average about ?20,000 each.

IR isn't ideal because skin absorbs that wavelength of light, meaning you don't get enough contrast. It's one reason why the actual in box kinect scans suck, as well as with photogrammetry. UV would be better but using UV flashes it dangerous for the skin and eyes.

Random image from the net:



Best,
Lee

[Wishgranter]

Guys its nice to see test results like these, BUT. im highly recommend to save for consultation with Lee, he go throught all thinkable configurations of possible improvements in last few years. so you get a lot of real info on what is good, what bad, and as im say in few earlier cooments, its worth all that money !!!!

From what im see in forum and few scanndata, results are not wery well, and think on two constants, time + money :-) you will try going to buy devices that cannot do what you intent, because you lack the knowlege and stuff around it. If im can say, trying build a fullbody scanner you waste aprox 10.000-20.000 EUR on not usefful hw ( sw ) and your results will be poor = more money, more time, more distraction, time-to-market and etc......

Lee can be seen as the mentor of fullbody scans, he has a lot of knowlege and understanding of all nescesary hw stuff, its a very complex stuff, one of the hardest.....


Im have earlier  discussing few things with Lee, and just from that SMALL info im get a lot of improvements on my data, not to say about the datast what im get from him ( my client was there for capturing photos ). And im cannot and want not give that info, just under some special things as we get scann data of Stewe Wozniak and few other VIP people, that was onetime exception because it wil not hapen anytime soon again........ 


 

[mala]

Have to agree with Lee on the use of IR, whatever method you use to project the IR pattern you still come up against the problem of skin absorbing IR well and hence giving a you a blurry result when looking at fine detail.
You are better off just having more hi res images(cams) with a bright evenly diffused lighting scheme.

UV is possibly of interest but i think mainly for motion capture, think using a UV make up diluted down and speckle sprayed on the face then illuminate with "disco style" UV lights or UV canon as we call them http://www.soundsavers.com/shop/all-lighting/special-effects-lighting/uv---blacklight/products/acme-400w-u-v--cannon.html?utm_source=google&utm_campaign=Google+Shopping+Results&utm_medium=Product+Feed&gclid=COT1vvehk7gCFSXMtAodmloAAg
Used in small doses these lights are perfectly safe they are after all certified for disco/show use.

At the end of the day I still think you're better off with just more cameras and white light.

One little trick that may also help to retrieve finer detail is to use just the blue element of your RGB images for processing mesh build, as skin absorbs red light well but reflects Green a bit and even more so Blue light.
The blue wavelength therefore brings out sharper detail because it hasn't been rattled around inside the skin whist subject to subsurface scattering unlike the red light.


cheers,
mala

[Infinite]

BTW Magnus, you're already getting very good results! keep experimenting.

The more experiments and different cameras positions and settings you try the more you will know what works for you. As there is no manual for this, no real right or wrong way. Lots of options to get different types of results