How accurate is your scanner ? The plane test.

[Hugues]

Hi guys,

For the past 2 weeks I'm trying to scan a part on my bike, kind of circle of 360mm diameter. I cannot scan it in one shot so i pan the scanner around following the circumference of the circle. But whatever i do, i end up with a mesh that has defect, i can see the circumference of the circle from one scan to the next does not mesh nicely, it's off by 1 or 2 mm. I tried to align the point clouds myself using Meshlab, it's a little better but still off by more than 1 mm.

I was thinking tonight that maybe the scanner does not read the surface properly. So i devised a simple test to check this. Results are below, the scan can be off by as much as 6 mm in some places !!!

Here is the test I've done:

- put the scanner in front of a perfectly flat white table, vertically standing, about 50 cm away.
- i checked that the scanner is leveled, and the table is vertical
- i made sure the scanner is focused, the distance i find with the sharpest cross is 50cm
- i scan the table, i should normally get a perfectly flat mesh, within 0.1mm.

But this is not what i get. The image below shows the deviation compared to a plane. I used Cloudcompare for this. I cannot get the scale to show on the right side. But i measured the worst deviation (red) at around 6-7 mm. You see the worst is in the upper left and bottom left corners.

You don't have to use any software to do this test, just scan a flat surface and check the result in Einscan. If you look at the scan from the side, you should be able to see if it's straight or not.

Would appreciate if someone can do the test so we can compare our results. Maybe it has to do with calibration. Well, hopefully. Because if all my scans are like this, this is not so good.

Maybe a workaround is to always delete these 2 corners at the end of each scan.

I tried to see if the problem gets worst when scanning at an angle, but i could not see difference, the same 2 corners are off. I wish there is a way to remove this defect by some kind of calibration.

Hope someone at Shining 3d reads this post.

plane_test.jpg

[24c]

Hugues,

Is this not a lens issue? I only say this because it seems analogous to SLR camera lens distortion.

I've scanned a matchbox sized object that was very square edged and flat, so is not the problem you are doing too large an area. If the surface doesn't have any reference points I can understand it getting confused, and usually this results in a poor texture...but that wouldn't be 6-7mm. interesting post though, and I will do a test, but scanner is offline, as trying to sort out Parallels activation on new iMac... Display Preferences, and projector working OK, and I used the Mac software to calibrate the projector colours, so I hopeful that it will be an improvement, when Parallels uses the native Mac software!

Display Preferences.jpg

Mike

[Hugues]

Hugues,

Is this not a lens issue? I only say this because it seems analogous to SLR camera lens distortion.

I've scanned a matchbox sized object that was very square edged and flat, so is not the problem you are doing too large an area. If the surface doesn't have any reference points I can understand it getting confused, and usually this results in a poor texture...but that wouldn't be 6-7mm...

Mike

Could be a lens issue.

Not sure about the reference point though. My test above is with one scan only. The software needs reference points to align 2 scans, but i don't think it needs them for a specific scan. Actually, it got its own black and white stripes.

This morning i made a long 2 hours scan, carefully cutting each of these curled corners ( i could see the curling on most scan actually now that I'm aware of it), it got pretty good, until the moment i click complete and the software crashed on me before finishing the mesh . What a ç%/%ç problem this is. I seriously hope Shining3d is adressing this problem. Lost my morning on this. Frustrating.

[24c]

Hi again Hugues,

After I posted the lens idea, I wondered whether if this is the case for all the scanners, then maybe a selection or lasso option might be a good idea, so you can restrict the viewpoint of the scanner? It would get around your problem. Another idea, the projector could project a 1024x768 pattern, but the left and right cameras only store a smaller centred window, say 800x600 by default to get around the problem, or give a High Accuracy as opposed to High Resolution scan.

it got pretty good, until the moment i click complete and the software crashed on me before finishing the mesh . What a ç%/%ç problem this is.

This has happened to me 60 Free Scans in 60 minutes, and the same, which is why I asked for a File>New Project function and a Save & Open dialog too...then if the software crashes (assuming you have saved) then at least you have something left. At the moment when it does this you have nothing but lost time

Feel your pain
Mike

[Eric_Wang]

Hi mike:

Thanks for your suggestion !

We will add New and Open project functions in the free model , the new version will be released in September 18.

Best regards !

Eric

[24c]

We will add New and Open Project functions in the free model, the new version will be released on September 18.

That's great news
Thanks very much
Mike

[Hugues]

Bump.

Anybody had time to run the test with their scanner ?

Would be important for me to know if I'm the only one with this problem. Shining3d is looking into it.

Just takes 5 min next time you scan something, point your scanner straight at a wall or any flat surface and check if you get a flat scan

Thanks

[Rebecca]

Hey Guys,
We did two tests just now
1. Plane test as Hugues did. Put a flat panel onto turntable, scanning under free scan mode for a single scan, and save .acs file. Please see the following result we measured in a 3rd party software, curled corners were cutting. you will find the result is within 0.1mm
flatness.jpg palne.jpg
2. As you know, when we talk about the accuracy, we mentioned about 0.1mm for a single shot data. To prove it, we did a measurement on a certificated ceramic ball bar with its centre distance of 63.253mm. The test result distance is 63.214mm, so the deviation comes out to be 0.039mm
ceramic ball bar.jpg ball.jpg

@Hugues, did you recalibration the scanner, and do the plane test again? With any questions, please contact us directly, you have Alfred mail, right? We are happy to work out the solution together with you.

[24c]

Hi Rebecca,
Interesting results, "scanning under free scan mode for a single scan" ... does this mean you used the turntable mode or moved the scanner? It suggests you moved the scanner.
It might be interesting if you auto meshed the flat plate and see what deviation would you get then? Just curious.
I can see this in the CAD software, because if I scan a cylindrical object, I can draw a solid and superimpose it over the meshed image, and see the texture variations on screen.
What 3rd party software did you use?

Mike

[manish]

hi rebecca,
single scan tolerance is within 0.1mm.
then why so we get whole scan deviation more than 0.8mm, sometimes to 1.2mm.
Regards
Manish