depth cameras, quality, price & openness

hey all

We’re currently struggling to get a 3d scan of a face in the shortest time possible. So far we played around with multiple primesense and kinects and stitch the clouds together to fill holes and dead angles. But those are just a little too inaccurate for our purpose (I think depth accuracy is around 3,4 mms). Also there is the problem with invisible beards. We were thinking about checking out one of those http://www.softkinetic.com/Solutions/DepthSensecameras.aspx
Not sure if they are more accurate but they do provide a shorter range… Anyone played around with these devices? We would also consider getting something more advanced but all scanners I found have proprietary protocols and rely and the software they ship with. That is something we want to avoid if possible.

Do you have any other thoughts, tips and hints?

thanks

hmm I think the laser based approaches might work better. ( sweep a laser beam across the face and map the displacement )

eg: http://www.makerscanner.com/

thanks. did you ever try it? I wonder how long it takes to scan the face. Also is the laser save? because this is going to be in a public environment, we can’t really guarantee that everyone has their eyes shut and keeps still for a few seconds.

hmm I think it takes at least 4-8 seconds.
I had to be still for mine.

no idea about safety - I think because the beam is dispersed over a larger area you might be okay.
but don’t know for sure.

this is what a high quality TOF camera can do:
http://www.muonics.net/cvMovies/3D-005-closeUpFace.mov

similar to the swiss ranger - its res is small (160x120), but the depth image is fairly noise free.
you could combine it with a SLR for some good RGB+D

also Panasonic D-Imager ( TOF )
http://www.youtube.com/watch?v=w34fO2AuzpI

thanks.

that is indeed rather noise free. Resolution would be high enough if the source is good. I will check it out.

have you tried using several photo cameras from different angles and shot in sync? or one camera with a static object?
the you rebuild the 3d model with all the images. It gives a quite nice result. Much higher res than the kinect, noise free and much better textures.

im on the same boat…

also i was looking at reconstructme skd, wich looks nice but its expensive an only for pc. Primesense has a new camera models (carmine) wich seems to have better resolution and shorter view distances, in the reconstructme blog are a few scans using this camera and a pair of glasses… they get a greaat result.

roy what do you means by rebuild the 3d model? how?

a while back i saw a demo of this 3d scanner.
http://www.artec3d.com/3d-scanners-for-professionals/artec-mht/- costs about 15K
http://www.creaform3d.com/en- costs about 10K

but i don’t think there is a way to interface it with OF. it’s more like a fancy photo camera.

@pandereto,
using software like Autodesk’s 123D Catch or this http://www.canoma.com/

Whow!!! that prices are absolutelly out of my budget, also as you i think that they comes with propietary software.

Im curious about something like the autodesk 123 catch approach, and this one “shapeshot” seems to do something similar and in their site they talks about an api… but i have not found more info, so when i read the roy response… maybe there is a library or something to do that. Shapeshot seems to be the soft that its used in the 3d photobooth at makerbot and the 3dea store in ny.

the best thing that i have seen for now, seems the reconstructme sdk it allows to use several sensors to reconstruct a single volumen, and i think that its factible to integrate in oF… but only for pc, and i dont ask about a full license price but for 600€ you have a license restricted for use one computer.

im still looking for a good way

P.D. thanks roy i will look at canoma

take a look at this.
http://p3d.in/WO2lq/shadeless
it was done by a local artist that I know. He built a 360 rig with a lot of point and shoot cameras.
I remember that he was using an open source software to compute the 3d model, but I can’t recall the name. I’ll ask him and let you know.

on the ohter hand, reconstruct me will tend to loose detail for each new frame it uses so youll end up with a model that is more or less noise less but with not very much detail with not good textures.

that loooks great!!!

It wil be great if you can get the name of the app…, i recentlly buy an ultimaker 3d printer so im looking for agood way to scan people, the multiple point and shoot cameras seems a good aproach.

I also take a look at skanect software

I think opencv stereo reconstruction is worth looking into: http://youtu.be/EfJDLhyeLy8

–8

try autodesk’s 123D catch.

Not sure if you are aware of this, but the official microsoft kinect sdk 1.7 now contains an implementation of kinect fusion and i believe this reconstructs 3D surfaces with high detail in real time (using GPU).

http://research.microsoft.com/en-us/projects/surfacerecon/