[Richard] just posted an Instructable on his ridiculously cool 39 Pi 3D Scanner! That’s right. 39 individual Raspberry Pies with camera modules.
But why? Well, [Richard] loves 3D printing, Arduinos, Raspberry pies, and his kids. He wanted to make some 3D models of his kids (because pictures are so last century), so he started looking into 3D scanners. Unfortunately almost all designs he found require the subject to sit still for a while — something his 2-year old is not a fan of. So he started pondering a way to take all the pictures in one go, to give him the ability to generate 3D models on the fly — without the wait.
He originally looked at buying 39 cheap digital cameras, but didn’t want to have all the images on separate SD cards, as it would be rather tedious to extract all the images. Using the Raspberries on the other hand, he can grab them all off a network. So he set off to build a very awesome (and somewhat expensive) life-size 3D scanning booth. Full details are available on his blog at www.pi3dscan.com
Stick around after the break to see it in action at Maker Faire Groningen 2013!
Why not 1 Raspberry Pi, 39 USB cameras, and a couple of hubs?
cause that would suck! And be a lot more expencive!
Why would it suck? And why would 40 USB cameras be more expensive than 40 cameras plus 40 Raspberry Pis?
Using heaps of RPIs is “ubercool” and using hubs is “boring” technical solution.
Welcome to the internets :-/
> “Using heaps of RPIs is “ubercool” and using hubs is “boring” technical solution.”
Apparently so is being clueless. The webcam ‘solution’ isn’t capable of doing what this person did with 40 RPIs.
Webcams is like giving birth to a guinea pig through your ear. No more like god luck knowing which camera is where. And a decent comparable webcam is in the same price range as the hole setup whit the pi system.
Raspberrypi plus camera is about EUR 50,- You can get a decent USB camera for that money. But it won’t be 5 MPixel. And a raspberry pi cannot handle more than one 5 Mpixel camera anyway. The USB on the raspberry pi is a bit mediocre in the performance field.
Now when you have 39 raspberry pi’s that are more or less identical, take a picture when commanded, compress it and send it over to a server (pi or not), you can achieve what this project aims for…
Actually, I’d consider taking 2Mpixel videos, synchronizing them by triggering a flash say about one second into each video, and then you have data to extract “moving 3D”…
A single USB web camera tends to saturate the USB bus on a Raspberry Pi.
it is just garbing stills, no live video
And you want all the stills at the exact same moment… So… not possible with hubs…
Because you would *NEVER* get them to all fire at exactly the same time. Any movement would cause errors in the model.
Hi, using webcams, I would not be able to all at the exact same time take all the images, and the average webcam with 5m pixel sensor would really be much more expensive.
I think if he’d tried to use 39 usb cameras, the small delay between each image capture would end up skewing the final 3d image.
With 39 systems all waiting for a single command to do a capture, the skewing would be cut to a minimum and, with good software, could even be calibrated out.
It would be far cooler to have a GoPro that spun around the person in high FPS mode in a quick spiral, this just seems.. wasteful? But it is interesting none the less, and those boards can be used for other things down the way.
“Unfortunately almost all designs he found require the subject to sit still for a while — something his 2-year old is not a fan of.”
Apparently, according to the article, not an option.
In addition what Elrinn said, I think that would introduce motion blur. I suppose technically it should be possible to compensate with image processing, but I doubt it would be able to achieve the same quality.
Indeed, the gopro isn’t exactly 10000 FPS (if 10K would be enough in the first place)
This could in theory make 3d videos right? What is the shortest time needed from one shot to another? (store all on individual pi’s, then merge afterwards?).
Yes, the raspberry pi can also shoot video with this, but for normal 3d movies you just need 2 cameras.
That’s for stereoscopic movies, where you present different views to each eye but still only from one camera angle. With this you could theoretically record hologram-style 3d video that would be viewable from any angle.
Correct, that would be possible. I would require some significant storage :-) 39x 1080p video streams :-) Maybe if I turn the cameras outward instead of inward I can hook it up to my Oculus VR Helmet :-)
I was thinking of low fps 3d mesh data. Then some way of matching each mesh with the previous and do interpolation to save space (although that might incur some artifacts for fast moving stuff).
And interestingly, I also thought of my oculus for some interesting applications: Capture a scene and then be able to walk through it while it’s playing. Would make a fun detective game (something like detect the pick-pocket or who poisoned the soup?).
Another application would obviously be nsfw stuff..
Which is known as ‘bullet time’.
https://en.wikipedia.org/wiki/Bullet_time#Technology
That’s for stereoscopic video, which presents slightly different images to each eye but is still only viewable from one camera angle. This could theoretically be used to record hologram-style video that is viewable from any angle.
I believe this is called free viewpoint television, look it up, its quite interesting, but not yet a mature technology
Thanks for supplying the correct term, I’m definitely interested in this.
I remember seeing someone implementing this as a conference system using two kinects: http://www.youtube.com/watch?v=EvTXLpTrEFw
Now this could stir up excitement among the panicky news media: “3d printed children.”
And even more so if one was holding his/her hand up like a gun!
With this, soon my robot avatar army will be complete and I will be able to conquer this world from my couch as was always my plan…
Now pair this with this:
http://www.ibtimes.com/3d-printers-build-entire-houses-contour-craftings-aims-print-2500-square-foot-homes-20-hours-video
and I will be set for mass production!
This reminds me of that drone they now have that captures a many many megapixel image live that it can capture entire boroughs or a city from high altitude, and it does so by using a boatload of common camera sensors set in an array.
It sounds simple but it still requires some work to make it all go smoothly, even when you get army/spook money to develop it. You need to combine all the sensors, combine the data, and then stream the whole selected region.