Hi all, I hope you can help me. I am scanning bodies using 20 cameras and a turntable, and getting quite good results except for the fact that my models are thicker than the subjects if I measure them. All of the circumferences of the body are off by 5% pretty consistently, while the heights are perfect. I have great coverage and am averaging 120 photos per body. I know it isn’t the model moving because the surface is good. I am using canons with 50mm lenses all in landscape mode. Any one else seen this?
Hi andycce
Do you have some screenshot of the issue ? how you measure and compare the results ?
I never encountered anything like that.
5% of lets say 30-40cm is something like 1.5-2cm.
That seems pretty hard to pin down on a soft body…
There really isn’t much you can tell from a screengrab of the forms, just that they are complete and should be accurate. When printing them out, however and measuring the forms they are accurate in every height measurement, but when compared with the subjects, all of the circumferences are over by about 5%. I am using these forms to make custom wardrobe, and it’s a lot of work to correct each one, and I would like them to be accurate right out of the gate.
Ah, so you print them off 1:1 and then you use a tape to measure the circumference?
I really cannot imagine how a model should behave like that other than massive alignment errors and that you should be able to see or rather there would be dozens of components.
I do have alignment issues with my models often enough but still the overall measurements are usually on the spot in every dimension.
Could it be an issue in the printing process?
Hi andycce
If you measure printed vs 3D scanned model then im absolutely not surprised that you see difference
Every single material have its temperature (stretching ) + physical properties so its absolutely OK that you see difference !
This is very well know issue of REAL world materials and just very few INDUSTRY application ( they cost 10-20+k EUR per license ) can solve the material stretching ( profiles per material used for printing - technology used ) issues.
Take a look on http://www.materialise.com/en/software/magics
I would readily blame the 3D carving for the problem as well if it were not for a few things. It is foam, that is already cured and dimensionally stable, carved on a well calibrated industrial 5 axis mill, all of the vertical dimensions are correct, and i have dozens of other things carved on this machine from cad data without any appreciable measurement inaccuracy. It’s only the things that are sourced from this scan data.
It just seems like the software reads planar data extremely well, but if you present it with data in the form of a column like structure, it isn’t exactly sure how thick the column is with respect to the height.
And the data is good, I only wind up with a single chunk after processing without having to do any manual align.
Hmm, that is weird.
But I really cannot see how it could be entirely up to the software.
I am scanning buildings and have in one very large case also plans for comparison.
It’s 100% accurate in every dimension!
And single objects like you have are the bread and butter of this kind of software.
I can understand that it is frustrating though.
How is the rest of your pipeline?
Could there some import option at fault, say scale x/y/z 1.05/1.05 /1.00?
I’m exploring everything, camera orientation, subject stability, machine accuracy, etc. I have known reference cubes in the scene that I use to calibrate as well as scale bar markers. It’s strange for sure. Was wondering if I wasn’t doing something with lens aberration that could help, but they are all 50mm primes with the object centered. Thanks for looking.
So are the measurements of the markers correct then?
what type of reference distance are you using, have you a screen shot of your cp’s and distance, it should show you a before and after distance so you can see what sort of accuracy you are getting, if your using a very short base line and it is slightly wrong it will make a huge difference
Yes reference cubes and all vertical dimensions come out accurate, final models are a little bit thicker when I take a tape measure to them versus reference measurements of the subject.
Even before I carved any replicas, my scans seemed a little “stout” when viewed on screen. I’m going to realign my cameras to see if I can’t help the software better understand the object, I was just curious if my problem wasn’t unique.
Thanks again.
Live subjects tend to never stay perfectly still. They can and do move ever so slightly. Even with 20 cameras and a turntable the photo sets you take will never match up perfect.
That inaccuracy between sets of 20 photos each may be why the models are bigger in all axes except the height.
Your reference cubes don’t move while the subject does.
Nice idea.
Breathing comes to mind.
On the other hand, he says ALL dimensions, even on arms and legs.
And if changes in position is the cause, shouldn’t there be as many narrower ones as there are thicker ones?
Andy, what happenes if you align only 20 images from one position?
In theorie there should be enough of an arm or so to be able to compare it to the distorted results.
Also the alignment report might be worth looking at, if there are discrepancies.
Are the markers all on the periphery of the images?
Have you tried an inanimate object or empty scene with markers only (also in the center)?
Thank you everyone for your advice and interest.
It is true that I would expect to get some deviation from subject movement, but I get the same anomaly all over the figures, even at the ankles, where I would expect to get less since they are effectively locked to the floor. And it is true, I don’t get thinner elements, only thicker ones.
If I take a single block of 20 photos, it looks good, but there is no way to output and check circumferences of a partial scan.
I have alignment and reference marks placed all around the figures, on the floor, vertically just to the side of the figure and actually placed on the figures.
I have not tried an inanimate object as a test as each replica costs $5000 and I would rather solve this in practical usage where I can utilize the results, flawed or not.
Gotz, I was thinking about your comment about buildings, which are an assemblage of planes, which may not give you the same issues, I wonder if someone were scanning columns, would they find the same thing I am.
I was just curious if the people who were doing those little figurine outputs ever noticed something, I don’t know if many of us are carving full sized replicas of their scans, and measuring the results.
Hi andycce
Highly recommend to look for “reference” measurement best done with properly calibrated laser scann. So you can better look for the actual size shift, where its come there. as 5% difference is too big, so there must be explanation why its happening. Its possible to setup a TeamViewer session so can check your project if the workflows you have used are good and etc ?