Holes in Mesh when using Laserscans - difference RealityCapture and RealityScan

Hi there,
I’ve experienced some issues durin mesh creation when using Laserscans only. I’ve compared Realitycapture 1.5 with RealityScan and I can see huge differences.
The first image shows the mesh in RC 1.5 - you can see, there’s one big hole in the front. I checked the laserscan data and there should be enough points in this area, the registration is also fine.
The last image shows the same data used with RealityScan. It has even more holes.
What can I do about it?
I also attached my current settings for mesh creation.
There is also a very big difference in the quality of the mesh between RealityCaptue and RealityScan. As you can see in the last image, there are these strange polygons between the table and the seats. There aren’t there in version 1.5


Hello @Fototechnik
As you can see there: RealityScan 2.0 Release Notes | Tutorial, there weren’t any changes in the meshing process between RealityCapture and RealityScan.
Are you using the same alignment/component in both versions? Or can you try to open created in 1.5 version in 2.0 version and process the mesh there? What are you laser scan import settings? What kind of scanner are you using?
Do you know the intensity range of the scan data? Can you keep the Minimal distance between two vertices setting for Mesh calculation to the custom one?

Hi @OndrejTrhan,
We are using Zoller&Fröhlich laserscanners. The files are exported as .e57 and the intensity ranges from 0 to ~5000000. In our preprocessing, we filter the values lower than 5000. We import the data with the following settings:

I’m simply importing the laserscans, click on “Align” and calculate the mesh. That’s what I do on both versions. Now I tried loading the 1.5 component/registration in 2.0 and the mesh has less holes than before, but there are still holes there.

I don’t understand where these holes are coming from? There are definetly points there in my laserscan (see image of pointcloud).

What value should I set for “Minimal distance between two vertices”? I tried with 0.000 and 0.001 in RealityScan and I see no difference. It looks strange with both.

Hi @OndrejTrhan
Now I tried again importing the component from 1.5 to RealityScan and it looks again strange:

The holes are still there and the strange polygons connecting the table with the chairs are also there.
I tried with both 0.000 and 0.001 for the “Minimal distance between two vertices” value but there’s no difference. I checked the settings for RC and RS and they are the same.

Hi, the problem could be in the used intensity values, as RealityCapture used the range 0 - 1 .
Would it be possible to share your dataset with us for internal testing? If so, I’ll send you the invitation for the data upload.

Sure, that would be possible. You can send the link.
I just saw that I also have the same problem in RealiyCapture 1.5 as well (but at another area of the scan).
And we’ve always used those intesity values, we haven’t changed anything in our preprocessing in the last years. But yes, we often had issues especially on tables, walls and flat surfaces with the holes.

OK, the invitation was sent to your email.
Thank you for sharing the data!

Files have been uploaded!
Thank you for help! I hope you’ll find a solution!

Thank you for your data.
I was checking this and it looks like the data are somehow interfering themselves. When there is only one scan, the seats looks good, but when they are aligned, there are some issues.
Was that space scanned like the example scans? I would suggest to add some additive positions, like from the ground, from the tables, under the tables.
Also, as these are Z&F scans, have you tried to import .zfs files?

Anyway, as there is this interference, I will create a bug for it.

Thank you for your reply!
I tried importing the .zfs files but in RealityScan this didn’t work, I got the error “invalid or corrupted input data”. However it worked in RealityCapture 1.5 but the mesh looked the same - very bad.

Unfortunately the whole space was scanned “only” from above, we didn’t scan below the tables. we have of course multiple scans all over the place, so there should be some points below the tables.

I tried normalising the scan data to [0,1] but that didn’t help either.

Is there any workaround that we could use to get a better mesh? Or do I need to manually edit it?

HI, you are right, there was a bug in RealityScan 2.0 for ZFS scans, which will be solved in the next release.
I am not aware of any workaround for now. I’ve tried different settings and it haven’t helped.
I’ve also created a bug for this. Our developers will check that and I’ll let you know about their finding.
We are sorry for the inconveniences.