Page 1 of 1

Meshing an interior scan pointcloud

Posted: Mon Feb 17, 2020 11:01 am
by isjtar
Hi everyone,

I hope someone can help me...
We just demoed a GeoSLAM where I work and I got some really nice pointclouds, but I can't seem to remesh them properly. All tutorials I can find use Poisson remeshing, but I doubt this is the right method as it is an interior scan, not a closed object.
Would greatly appreciate some help, been stuck on this for days.

If anyone cares to take a look, you can find the scans here, the first one is the lightest.
https://www.dropbox.com/s/jhmc6eqsa5ekk ... 1.laz?dl=0
https://www.dropbox.com/s/bdlkfr84rtvdf ... 2.laz?dl=0



Thanks in advance,
Isjtar

Re: Meshing an interior scan pointcloud

Posted: Mon Feb 17, 2020 10:39 pm
by daniel
Ok, the noise is quite big, and meshing complicated structures without clean normals is always a bit hard.

Anyway, here is what I did:
1) first, remove the noise: I used the SOR filter (https://www.cloudcompare.org/doc/wiki/i ... SOR_filter - default parameters)
2) I computed the normals (Edit > Normals > Compute), with a radius of 0.05 and a preferred direction of '+Barycenter' (which means that most of the normals will point towards the barycenter which is approximately at the center of the room)
3) I eventually used the Poisson Reconstruction plugin (here with a level of 10 - the noisy parts are of course harder to reconstruct properly)
poisson_external.JPG
poisson_external.JPG (132.82 KiB) Viewed 5182 times
And here is what's inside (it's always the trickiest, especially since it's more noisy and it's hard to compute normals correctly)
poisson_internal.JPG
poisson_internal.JPG (221.09 KiB) Viewed 5182 times

Re: Meshing an interior scan pointcloud

Posted: Tue Feb 18, 2020 11:09 am
by isjtar
Thank you daniel, much appreciated.
So it seems the scan in itself is too noisy, very dissapointing fo such an expensive device.

Re: Meshing an interior scan pointcloud

Posted: Wed Feb 19, 2020 8:41 pm
by daniel
I would say it's good for big objects, walls, etc. And to extract global dimensions, trends,etc. But definitely not for small objects indeed.

There are probably also some tricks to improve the accuracy when scanning?