ICP Registration across ~15 meshes

[salamanders]

I'm a brand new user, so please holler at any misunderstandings. My steps (and question) are:

1. Load a folder full of 15 ASC files (depth-scans of my FACE) at once - success! And fast! And it even asked me if my 3 columns were XYZ! Nice!
2. Try out the "segment" tool to draw around my shoulders across all 15 - success!
3. Press the wrong button for "keep inside" vs "keep outside" - whoops. Can I control-Z? Doesn't seem like it. Reloading.
4. Keep-outside. success!
5. Oh! this creates two clouds inside each uh... group(?). So maybe it didn't matter which I clicked. Cool. Hide my shoulders by un-checking the right boxes. 15 un-checks. - success! (I think)

Now I want to line up all 15 scans and create a wonderfully accurate point cloud (of my FACE).

When I click all of them, I can't do Tools > Reg > ICP. Hum... Ok, it allows me when I click 2 clouds.

Lining up 2 arbitrary choices from the 15 - Success! (I think. Do I need to do something to say "apply these changes"?)

No, it did do the alignment, yay! But... hol'up. My nose isn't matching.

I guess I could do pairs, only takes 14 times to align everything to whichever one has me facing straight forwards.
But there isn't any sense of N^2/2 alignments. So everything has to click into place with that one center alignment. Huh again - I don't think my side-shots have enough in common with the center-shot, they need that "chain" that links them back to the middle.

My question: Is there an easier way to toss in a handful of scans, say "align however you can every which-way using whatever technique is best" and get back out some nice tight evenly distributed mesh?

And if not, what is the usual best way to get started?

Thanks for any advice! Great app.

[daniel]

Hi,

No sadly CC doesn't support N-to-N scans matching, it can only do pairs... And mind that for ICP to work correctly, they scans have to be already roughly registered.

Some people still try to match multiple scans, and indeed you have to apply the "chain" of alignments (2 with 1, 3 with 2, etc.). But this can lead to some big drift in the end.

[CloudyCloud]

I'm searching for an automatic multiple clouds registration since a few days, and I only found some commercial (unfortunately very expensive) apps offering such functionality. I would happily spend some money donating development of CloudCompare if that could help implementing this feature.

Do you guys know any software that does not cost a fortune and is capable of doing that? Just coarse matching would be enough as long as the output could be finely aligned in CC with ICP.

Any tips greatly appreciated.

[daniel]

I don't! But there might be some academic tool/algorithm available somewhere? Maybe even on github?

[CloudyCloud]

Thanks for the suggestion! I found one using a tool called 'Point Cloud Library'.
In the bottom of the page, 'Globally Consistent Registration of Multiple Point Clouds': https://prs.igp.ethz.ch/research/comple ... louds.html
Seems like a ready to use setup, but looking at the PCL documentation I think it will be very difficult for me to understand how to use it... At least at the moment I don't get how to start with it.

[WargodHernandez]

Thanks for the suggestion! I found one using a tool called 'Point Cloud Library'.
In the bottom of the page, 'Globally Consistent Registration of Multiple Point Clouds': https://prs.igp.ethz.ch/research/comple ... louds.html
Seems like a ready to use setup, but looking at the PCL documentation I think it will be very difficult for me to understand how to use it... At least at the moment I don't get how to start with it.

This looks interesting to me and in line with something I need to do anyways, I will take a look at turning the sample code they provide into a CloudCompare plugin, maybe just extend the current PCL CloudCompare plugin.

I'll post here if I do end up implementing it.

[CloudyCloud]

That would awesome. Good luck :)

[CloudyCloud]

Hey WargodHernandez,
Did you have a chance to look into it?