3D scanner for 300 dollars or less?

[Namielus]

I have been trying go make various rinky-dink camera based 3d scanners.
With varied success. The problem is it cant capture anything that moves. (Even a human trying to stand still is not enough)
123D catch works really well, if you do it right the result is incredible.
The problem is the object needs to be stationary. It does not work very well for humans since they move around.

http://www.123dapp.com/obj-Catch/Sigurdr/1043522

Therefore, I have been trying to think up a rig to set up multiple cameras in different angles. (Cheaply)
The goal is to arc some pvc-pipes over an area that will allow for a t-posed full grown human to be captured from multiple angles at once.


I have found some really cheap cameras on alibaba, down to 10 dollars (!) per camera if I order in bulk.
The quality is good enough both for generating the mesh and texture.



Since all cameras look exactly identical, I will mount a cardboard plate in front of them with a strong, defined and unique pattern covering everything but the lens. This is also to provide clear and definite reference points for manual correcting.
The cameras in the pvc-arc is not to scale, and was modelled from the specs of a miniature camera.

Connection wize, it seems to be a challenge of usb power, bandwidth,confused drivers and IRQ issues to connect 30+ cameras via USB.
So, that might not be feasible to trigger them all at once.

The cameras in that price range usually do not have any option for remote triggering, so I would have to modify it by soldering all the shutter buttons to one circuit/button that can complete all the individual shutter circuits at once.

My questions have not been answered very well by the manufacturers, nor do I really know how to properly ask how the shutter button circuits work.

My technical english vocabulary is very limited, I am basically wondering how to find out if the shutter buttons usually are simple 2-wire circuits that needs to be closed to trigger the shutter.

I imagine its not a half press button, but if it is, how hard will it be to solder a simpler trigger to all those cameras?

What I mean by half-press, is the cameras where gently pressing the button will focus and fully pressing down will capture a photo.


Does anybody have any good suggestions for the type of solution I can use to trigger cameras like this? The quality

[Likay]

It's a cool idea.
A different approach to the almost same idea is to have the object on a rotating board while filming with one or more cameras and then later analyse the movie and create a 3d-mesh. Needless to say that the object has to be still during the entire scan though...

Hardwiring cameras: I once made modifications to two cameras (cheap olympys models) when i twinned them into a stereocamera. If i'm not remember wrong i think i might even have posted some images of it on this forum. However: I did have problem with the sync between the cameras and i got a fair deal of shots where the cameras:
1: Choose different light/flash conditions etc and
2: Bad timing which leads to missed sync and ruined stereoimages.
My approach was hardwiring by soldering the ground, "halfway shoot" and the "shoot"-button with 3 wires between the cameras. The cameras i used didn't have multiplexed scan of the buttons so this was ok. In case the cameras are using multiplexed scan for the buttons you can use c-mos bilateral switches for the interface with advantage.

Canoncameras can usually be used with StereoDataMaker (CHDK software) which guarantees perfect sync (through the usb) but it can in odd case cause crashes (softwarecrashes which needs the batteries taken out before reset). Canons are not that cheap though. I use this solution for my nowaday twinned camera but in most cases it rests at home in favor for the w3 (worse pictures but easier to handle).

End of line: It's definitely possible to sync the cameras by hardwiring but i don't think you can expect a 100% sync so it might still be necessary with an object that doesn't move during the take. I'm talking about parts of a second but if the object moves it can be enough to ruin a shot.

[Namielus]

I have already made a rotating camera rig, that is big enough for a human, but it just doesnt work especially for t-pose.

The person simply sway around too much after a short period of time. If you spin the camera around too fast, it will have a jelly like skew because the sensor scans from top down and that distorts the image.


I am looking for cameras with a fixed focal length, different exposure values do not seem to be an issue really. (remember I am not talking about stereoscoping imaging, simply using the positional data of the combined photos to generate a mesh with texture)
Most important is that they all have consistent focal lengths.
They dont need to be synced within milliseconds tho, the person is still expected to stand still.

I am not talking about capturing a 3d model of someone in the middle of a jump etc so some minor sync issues in the millisecond range is okay.

[cybereality]

I think its a great idea. These things are called light-stages, they have been used by ATI to make this:


http://pc.watch.impress.co.jp/video/pcw ... 7.flv.html

The rig used to scan her face looked similar to this:


http://gl.ict.usc.edu/Research/DigitalEmily/

[Namielus]

Yea it looks exactly like what I am planning to make (except im making a dome) But its going to be a challenge finding the right price vs quality.


It has to be affordable and therefore 30x 100 dollars is too much at my current budget. Forget setting up 30 dslr's
My main concern is still how to trigger them all.

Here is a sample photo from the 10 dollar camera.


Note that the variations of light in this scene, cheap cmos sensors do not handle that well.
The lighting condition in my setup will be more even. I also think the guy that took these pictures for me took them all
through a window and this will degrade the overall image quality.

Its also going to be a pain in the ass transferring files from 30 memory cards, but the benefit is that once it is set up it only takes a few seconds to capture each person.

The 3d model is generated by the autodesk cloud service, and you can start multiple projects at once. The time for each
project varies betwen seconds and up to 30 minutes.

The model is extreme in detail and the topology is a mess, but there are auto retopo programs out there that work pretty well and will generate a model usable for 3d games.

Autorigging also exist, so there are plenty of tools for Average Joe to make a high quality textured avatar of himself.

[Likay]

My main concern is still how to trigger them all.

If you plan to hardwire the cameras you can open one up and check the electrical connections for focuslock, shoot, zooming, menu etc. If all these keys have one terminal in common (it's probably ground, check if it is to determine) then you should be able to hardwire them all together quite easily without extra equipment. Solder a 3-wire cable to each camera (ground/0v, focuslock and trigger) which then is connected in parallell with all the other cameras. One camera or external switches can be used to trigger them all.

If the keys are organized in an "array" where there aren't one common terminal for all the keys it means that the keypad have multiplexed scan. You can't hardwire these directly without encountering problems. The solution is easy but demands a little more work. A c-mos 4016 can be used (one ic has four channels but i think it's easier to use one ic/camera since they're cheap) with each camera to provide a solution. I'm fairly sure the ic has to be mounted close to the camera it's suppoused to control to avoid complications. If the internal voltage of the cameras is high enough you can use this to feed the extra ic but an external power supply might be a lot easier to do (a 9volt battery will feed all ic's without problems and you would probably also do the focus and triggerbuttons in the same case).
In case you use the internal voltage of the cameras a 3-wire cable is needed (ground/0v,focuslock and trigger). If you use an external powersupply then a 4-wire cable is needed (v+,ground/0v,focuslock and trigger). If it seems of interest i can make some kind of drawing.

[Namielus]

Likay thanks for a comprehensive answer.

Now I just have to find a chinese interpretor. The best answer I get from the manufacturer,
is "yes very good camera" so their english is very limited.

I could of course order a sample camera, but the shipment cost by fedex from china is prohibiting.
So I would like to figure out what camera I want before I order 30 of them.

[EdZ]

Do the cameras save data locally, or do they operate over USB? You could conceivably grab a bunch of RaspberryPi's, and plug one (or a few, depending on driver conflicts) cameras into each. Hook them all up to a switch by ethernet, and do your triggering that way.

Bonus: when the RaspberryPi camera add-on arrives, you already have most of a full VICON-style mocap setup ready to go (Pi's do blob discrimination and tracking onboard, and just send raw point locations to a central server) if you wanted to switch between capturing a mesh, then capturing animation data.

I could of course order a sample camera, but the shipment cost by fedex from china is prohibiting.

Alibaba has a subsidiary called AliExpress, which is set up in more of an ebay fashion and aimed at consumers rather than suppliers. If you can find the same camera at AliExpress, the shipping for a single unit will likely be significantly cheaper.

[Namielus]

I think the usb only function as a mass storage device. The camera does on board image processing and save to memory card.
Might be other alternatives tho.

[Nick3DvB]

Cool idea Namielus!

It might be worth looking at very cheap USB "endoscope" inspection cameras:

http://www.ebay.co.uk/itm/5M-USB-Waterp ... vi-content

They have led lights, long cables and you can get them for less than $10 now.

The RasPi camera module could be a good option, but you might be waiting a while for it to turn up.

You might not need as many cameras, if you use some kind of interpolation to fake intermediate views,

this is a bit old now but worth a watch: http://www.youtube.com/watch?v=WYezwsFfxvE

[rhinosix]

Now I just have to find a chinese interpretor. The best answer I get from the manufacturer,
is "yes very good camera" so their english is very limited.

Let me know if there's something in particular you need translated.

Also, I have one of those cameras in your picture if you want one to pull apart. I think it's a 910 IS (the model number varies by county).
You can load CHDK onto it. Though the LCD is kind of busted, and I can't find the battery charger at the moment...

How are you handling lighting?

Is this useful?: http://sabsik.com/Cam2Com/index.htm

Have you looked into some kind of laser scanner?