Instructables

123D Scanner - Home made 3D Scanner

FeaturedContest Winner
Picture of 123D Scanner - Home made 3D Scanner
Hey - have a look at my new project HERE

In this project I built a 3D Scanner, that enables generating 3D models of physical objects.

The files can later be viewed in 3D software (GLC Player, Sketchup, Rhino, or sites such as http://3dfile.io)
and even manipulated into .STL file and 3D printed.

The software for this project is completely free, I am using Autodesk's 123D catch, Link:123D catch
The 123D Catch is a great software, it requires taking many photos of an object all around it, and uploading it into the software, and it returns a 3D file.

Since I really liked the solution but did not wanted to take the photos myself - I built an instrument that does that -
description hence.

Please note that this document does not intend to explain how to use 123D catch (this can be found here)



A really nice scan can be seen interactively here:
 

.
 
Remove these adsRemove these ads by Signing Up

Step 1: Operation principle (can be skipped)

Picture of Operation principle (can be skipped)
(This part can be skipped, it explains the logics behind the structure, The scanner will wok without this part...)

The design of the scanner is quite simple, and consists of 3 main parts:
  1. Scanner chassis and stage
  2. Scanner arm (carries the camera)
  3. Scanning surface
Each parts has it roles -
  1. The chassis - this is the part that sits on the table and carries the scanning Surface and the scanned object on it
  2. Scanner arm - The arm has the camera in the end of it, and a servo motor that presses on the button and take all of the pictures, the arm rotates around the scanning surface and takes pictures every few degrees. Pay attention that there's a weight to balance the own weight of the camera (*)
  3. Scanning surface - a round board, with colored stripes on it, carries the scanned object. (**)
(*) The reason that the arm rotates around the object instead turning the object around is in purpose to make the stitching algorithm work easier, since if the object rotates, and the surrounding stays the same, the software work becomes complicated and the model doesnt work so well, environment -same as light sources shadows and all. (I did built one model with turned stage but the results were awful)
(**) The color stripes on the base are also in purpose to help the  stitching algorithm - the help (colors, number of lines, orientation) to stitch.
1-40 of 41Next »
stuffdone4 months ago

Most video editing software can output a series of stills from a video. Why not just make a turntable, mount video camera and turn it 360 degrees to capture video then extract the stills from that? No complicated electronics, just a simple turn table and video cam. For quality I have a Fuji SLR that can connect to my PC so I have optical zoom, macro etc.

I have not tried this because I just now read this article and this is what came to mind. If someone tries this before I get the chance, please post results here.

Us lazy people make the best inventors...we always look for an easier way !

Another thought. It would not be hard to use a lazy susan bearing to make a table that could support a person. With the simpler video method you could do whole body scans perhaps!

Ok...more thoughts. Video at 1 FPS ( time laps setting ), turn table once in 20 seconds. Export frames for 20 frames per 360 degree revolution. Experiment with different frame rates and RPM of table or course.

alluvian5 months ago

Love the idea, and it seems to produce pretty good results. I am torn about using catch or a laser scan solution. I probably want both, as I plan to tinker in games and 3d printing. Catch will give better game models since they are wonderfully textured, but I think laser solutions give better mesh/point clouds.

For lasers I would suggest looking into the moedls app for iphone, this thing looks amazing. www.moedls.com too bad the kickstarter failed. The system uses a green line laser and takes a video of the subject turning on a turntable.

I also love your idea to automate catch. In my head the rig I see is two rings of diffused led lights affixed to a turntable, and then rotate the object and the turntable with the lights in a dark room or cover the thing in a box maybe during the day? This is effectively MORE complex than your solution, but it sounds like a fun build to me. Reality states I will never get around to doing this though, heh. Just having the lights on the turntable in the darkness would speed up the picture taking process immensely though.

Has anyone tried 123d catch with NO shadows on an object? Does it need shadows to function correctly?

Nyxius10 months ago
Wouldn't it have been simpler to keep the camera stationary and to rotate the platform?
Armagedoom Nyxius10 months ago
+1
DanW1310 months ago
Very Cool, I wish I was smart enough to build on, I'm not a Tech Geek kind of Guy but it would take me months & lots of Q & A before I'd get to having a fully functional 3D scanner…I seen Scanners like this on line sell for $1,000's. Your cost was I'm sure far cheaper maybe a few $100 ? Nice Work.
soilwork1 year ago
the output can be used for a 3d print or cnc job?
wavegm (author)  soilwork1 year ago
You can export an OBJ file or STL
But you still need to make the corrections.
CrazeUK1 year ago
Hi, i am totally new to this. And thinking of trying your project.

Although i am thinking why not use a webcam? They are quite high quality, and because of the small field of view, it wouldn't need a high MP count?

Additionally, you can totally disassemble it, and only have the actual lens and sensor mounted greatly reducing the weight?
wavegm (author)  CrazeUK1 year ago
I did buy a webcam to try and do it - It might have been even better for the images to be on the hard drive instead of on the SD.
The thing is that the webcam wasn't so good - they dont have OPTICAL ZOOM that according to what i found out is the most important part.
the MegaPixel number doesn't matter much (i bought 5 MP webcam by HP that gave bad results and I used 2.3 MP fuji that gave good results) Now I use Lumix FX90 which is better for many reasons.

Share what you manage to do!
rlaganson11 year ago
Does it matter where the lines go?
wavegm (author)  rlaganson11 year ago
Anything new?

Have a look at http://www.keerbot.com
wavegm (author)  rlaganson11 year ago
Nope -
I think that small number of straight lines, colored,
will do the job
(what you do here is "helping" the stitching algorithm)

Do you build it ? I awnt to see !
This website (Wiki) is in German but you will get the picture ;)
http://wiki.fablab-muenchen.de/display/WIKI/3D+Scanner+Wissen+-+Scannen+mit+Autodesk+123D+Catch
Use translate.google if you need it.
It explains how set up a platform to rotate an object and get succesful results.
The secret is to have enough reference point on the objects mat as 123 uses them to align the pictures.
For my old setup I used some Lego blocks, and other stuff for reference but any good pattern should do.
At 3Dsom you can see a similar system used.
rondust1 year ago
I appologise in advance.

Why dont you just use an old record turn table??
Remove anything you do not need from it, centre pin, arm etc.
Mark out lines on cardboard and put onto turn table plater, mark the sides of the plater for consistant positions and put an alignment reference on the base as well..
Mount the camera in a fixed position, if you used a socket for the arm mount it could be unhooked and packed flat. No wobbles and rigid.

Put item on plater, manually take picture, manually turn to next position and aligh marks, take picture etc etc until done.

Cheap, compact, faster.
wavegm (author)  rondust1 year ago
please read the remarks
rondust wavegm1 year ago
I appreciate that if you dont pay attention to it, the lighting would be poor and yes shadows would affect the final product. I assume that would be one of those "get it right" thoughts when setting up.
A reflective bottom and side surface and good lighting would fix that easy and still be table top.

However if the idea was to ALSO capture the change of background as a movement and/or lighting from a particular direction casting shodows of the object, then you need to construct a supporting ring or surface with bearings to fix the wobbles on the camera, the bearings need to be set apart on the rail or surface and use a triangular frame to raise the camera - no wobbles and huge range of adjustment for camera down angle from bottom to top of the support triangle.
You could even tilt the top of the triangle support towards the object or away for other effects, and even use a servo motor to move the camera up and down while taking images.

My assumtion was that an item would be well lit from all angles to get the clearest image of it in 3D rotation - easier done by rotating the item infront of excellent lighting all around the item.

I did appologise in advance.
Please dont think I am putting your project down, your method has great merits for particular effects.
My understanding (from reading the stuff on the 123D Catch website), is that the shadows/lighting on the object are a part of how the software reads surface topography. The shadows/lights need to be consistent relative to the subject across the photo set, or else it gets confused and reads the topography incorrectly. Hence why the author's experiments with rotating the subject instead of the camera gave bad results.

If that's the case, it wouldn't be enough to rotate the subject under good lighting, you'd also have to rotate the lights in sync with the subject. That means either affixing small lights to the turntable (easy, but probably not the best kind of lighting for this), or putting your studio lights on computer controlled motorized circular dolly tracks (completely impractical).

The fact that the process can work using hand held camera (as per the demos on the 123D Catch site) means you don't really need to worry quite that much about stability with your camera rig. As long as the pictures are sharp and the distance & focal length constant, you probably get more than enough precision for an accurate model. I'd guess that a little bit of camera wobble between frames probably won't make an appreciable difference, otherwise handheld would be completely useless.

That said, some kind of sliding mount on the vertical bar of the camera arm so you can take additional series from other elevations would definitely be good. and a circular bearing or track would be better for getting true 360 degree coverage (this rig only gets maybe 330 because the stand prevents the camera arm from moving completely around).
wavegm (author)  rondust1 year ago
I think I get your point - can you sketch it ?
I though of trying to change the background but it is complicated technically
oldboffin1 year ago
"he said I should write a patent on it..."

I think putting your idea on instructables might be disclosure :-)

( you can't patent it after you have disclosed the idea without a non-disclosure agreement )
wavegm (author)  oldboffin1 year ago
I don't really think a patent on that idea will work,
I lack the will and the funding for messing with that ...
Theres a school of thought that reckons a patent is a waste of time, as you spend months keeping your invention secret as you develop it, then when you have your patent, you walk the streets showing it to potential customers for months :-)
Plus there are some countries that don't give a hoot about intellectual property !

I love the end result of your photoshoot with the pegs !
4363531 year ago
Why not rotate the base, rather than the camera arm. I think it would be more stable. Nice work though.
436353 4363531 year ago
Never mind. Should have read other comments.
For some set ups it might be useful to turn the camera around the object, but from my experience I can say that a rotating table with some markings in different colors work well and would keep the setup a bit more compact.
Also most modern cameras offer a remote trigger, eiter using a wired connection or a remot control - might be worth of using it for a more permanent solution.
One of the biggest problems I encountered so far is the simple problem that not all areas can be made visible to cam from one fixed height.
I was experimenting with a dual cam setup before I started laserscanning.
During that time I noticed that it often makes sense to use a low resolution to build an object and a high resolution to grab the texture of an object where needed.
Sadly I did not take any pics but I try to describe the rig I used as I think it will make you system easier to replicate:
For the object table I used simple MDF board about 60x60cm.
In the centered I glued two A4 sheets containing random lines and patterns in different colors to give the programm something more to work with.
The main difference to your setup is that I added two boards on each side with a cross section ot top, creating a frame over the center object area with the cross bar about 50cm over the object table.
In the centre ot that cross bar I added a 10x10cm and 1cm thick steel plate and drilled a 8mm hole through the metal and the cross bar.
For the turning mechanism I used a simple bike rim with one side of the axle going through the cross bar and steel plate.
Being a steel rim it was easy to weld a proper mount for cams on it and it was stable enough to turn without dropping on the cameras side - but adding a counterweight on the opposite side is better.
I turned the rim by hand and had simple marks on it or every 20° to line up with the cross bar, the trigger was activated using a remote control.
Whole setup took about 3 hours to built and worked fine with 2 cameras attached to it.
If I would biuld it again I would use one cam on each side instead both on the same to make sure the weight is distributed more evenly.

Of course it is just a suggestion and you should decide on the frame/plate dimensions depending on the rim and object size you intent to use.
wavegm (author)  3D_Laserfreak1 year ago
Thanks for the detailed reply,
any drawings/pics would help for sure to understand ...

What cameras do you use ? I searched a lot for remote controlled cameras in a reasonable price and found nothing

I even thought of building an android app for that...

jrd2101 year ago
Great idea, but a more detailed parts list would be helpful to those of us living a long way from stores and are you saying MicroRax is the only way to go--still need parts list please.
wavegm (author)  jrd2101 year ago
have a look at the pictures -
I think that you can make it much cheaper then the microrax - wooden platform or any other bars will work.

Improvise !
sarveshk1 year ago
Great!

I have also tried similar thing. I added different pattern for background. But that did not worked as expected. For me, 123D catch gave better results when I used abstract backgrounds. I have also read somewhere about a user's experience, which was similar to mine.

However your project came out pretty well! Now I'm gonna try with your setup!

Thank you for the great project!
wavegm (author)  sarveshk1 year ago
I am glad you liked it - please share whenever you have anything new I would like to see!

if ou have any questions - feel free.

Gilad.
Nice. I would rather make the object turn other than the camera.
Good job. I love your idea

If you rotated the object, the lighting would change relative to the object and the shadows would move ruining the 3D.
wavegm (author)  leo_p1 year ago
you are right - I even had the re-approved by Autodesk in the 123D facebook page
Because of the way 123D Catch works, you usually get better results by moving the camera.
Jayefuu ___1 year ago
+1, the results are terrible if you turn the object not the camera.
That's one wobbly camera arm! If you could stabilize it better, you would create your images much faster. How long currently does it take to do a full "scan?"
wavegm (author)  captain Jack1 year ago
The arm is quite flexible that I know - any change into it will probobly result increase in the weight, That will make the rotation servo work harder.

I let it rest for 1 sec and it is enough.
the complete scan takes less then 2 minutes.
rimar20001 year ago
Interesting. I think would be easier to turn the object instead of the cam. You would avoid the trembling and end of displacement.
wavegm (author)  rimar20001 year ago
The method of turning the camera certainly gives better 3D model,
the down is the complicated mechanism that needs to be built.
1-40 of 41Next »