Deformation analysis statistics

Discuss Leica Cyclone 3DR software here.
StuEv
I have made 50-60 posts
I have made 50-60 posts
Posts: 52
Joined: Mon Nov 16, 2015 12:52 am
8
Full Name: Julian
Company Details: University
Company Position Title: Student
Country: Germany
Linkedin Profile: No

Re: Deformation analysis statistics

Post by StuEv »

the second point cloud with 6mm to 60mm point spacing has 700000 points
User avatar
gilles_3DR
V.I.P Member
V.I.P Member
Posts: 235
Joined: Mon Dec 08, 2014 11:41 am
9
Full Name: Gilles Monnier
Company Details: Hexagon
Company Position Title: 3DR Technical Manager
Country: France
Linkedin Profile: Yes
Has thanked: 2 times
Been thanked: 28 times

Re: Deformation analysis statistics

Post by gilles_3DR »

ok, I understand why you export with a mean space of 1cm now!

Then, I think you are creating the mesh in the right way: in the second step, the mesh is really fitted in the middle of the point cloud thickness and is therefore a good reference for the next step...

When you send your request for a full license, please mention this thread in your request!
StuEv
I have made 50-60 posts
I have made 50-60 posts
Posts: 52
Joined: Mon Nov 16, 2015 12:52 am
8
Full Name: Julian
Company Details: University
Company Position Title: Student
Country: Germany
Linkedin Profile: No

Re: Deformation analysis statistics

Post by StuEv »

thanks,
and what would you recommend for the second point cloud? in the middle i have 6mm spacing and left and right ends of the wall i got 60mm spacing and therefore i am getting holes in this mesh.

and second question: is it possible to make a comparison like in geomagic of the deviation of a mesh and a pointcloud like this:
Geomagic12.png
You do not have the required permissions to view the files attached to this post.
User avatar
gilles_3DR
V.I.P Member
V.I.P Member
Posts: 235
Joined: Mon Dec 08, 2014 11:41 am
9
Full Name: Gilles Monnier
Company Details: Hexagon
Company Position Title: 3DR Technical Manager
Country: France
Linkedin Profile: Yes
Has thanked: 2 times
Been thanked: 28 times

Re: Deformation analysis statistics

Post by gilles_3DR »

for the 2nd point cloud, I wouldn't recommend to create a mesh out of the point cloud: I would use the first as a mesh and the 2nd as a cloud to do the inspection...

The difference I see in the pictures is that you only want to view a slice of the inspection.
Is that correct?

If so, the only way I see for now is to create a plan perpendicular to your wall and then use this plane to keep only the points which are at a given distance...
In the coming version, we have integrated a clipping plane tool which will allow to visualize only a slice of objects...
User avatar
danielgm
V.I.P Member
V.I.P Member
Posts: 195
Joined: Thu Jun 20, 2013 10:45 am
10
Full Name: Daniel Girardeau-Montaut
Company Details: CloudCompare
Company Position Title: Administrator
Country: France
Linkedin Profile: Yes
Location: Grenoble, France
Has thanked: 11 times
Been thanked: 34 times
Contact:

Re: Deformation analysis statistics

Post by danielgm »

Hi guys,

Just to add my two cents in :)

If you really want to estimate the accuracy of a sensor then it's a bit weird to use the same sensor to produce the "ground truth" data and the compared data. But maybe you are more interested in the repeatability of the measurements?

Anyway, you have put your finger on one of the most problematic issue when preforming 3D data comparison with points clouds and/or meshes:
  • as Gilles pointed out, if you want to use a cloud as reference you'll need a high density to get reliable results. I've shown during my PhD (the manuscript is in French, sorry) that the error induced by the reference cloud density is bounded. And you can use local modelling strategies (e.g. extracting the 'mean' local plane) to get a better estimation of the distance to the real surface in presence of noise or small holes. This approach can give locally some weird measurements, but the global statistics are valid and reliable.
  • And if you use a mesh then you must be aware that the meshing method will also induce a bias depending on the meshing algorithm you use. Same thing for the eventual hole filling process (in the general case it's hard to reproduce the real surface roughness when filling holes). I don't believe any meshing method is 100% reliable especially when applied on real life scans. Therefore once again you may locally have some unreliable measurements, but the global statistics are good.
This said, I agree with Gilles that if you are not confident with your cloud data AND if you have a good mesher then it's generally a better idea to use the mesh as reference (and I have heard that 3D Reshaper does this very well ;) ).

And last but not least: if you are interested in extracting (global) statistics to estimate the accuracy (or the repeatability) of a scanner, then it may be to a good idea to scan a flat surface (your wall can be a good candidate if you are confident in its flatness) and then compute the deviations of the points relatively to the best fitting plane. This way you avoid the influence of both the cloud density and the meshing algorithm.

Of course you'll need a good confidence in the surface flatness... But once again it's always the case when dealing with accuracy estimation. If needed you can read very interesting things in the papers of Derek Litchi for instance (https://scholar.google.fr/scholar?q=der ... _sdt=0%2C5).
Daniel
CloudCompare admin
http://www.cloudcompare.org/
User avatar
3DForensics
Honorary Member
Honorary Member
Posts: 1979
Joined: Mon Aug 03, 2009 1:52 am
14
Full Name: Eugene Liscio
Company Details: AI2-3D Forensics
Company Position Title: Owner
Skype Name: eliscio
Location: Toronto, Canada
Has thanked: 13 times
Been thanked: 70 times
Contact:

Re: Deformation analysis statistics

Post by 3DForensics »

...for the reasons stated above, I was suggesting to use two dense point clouds. As pointed out, there can be problems either way, but if the point clouds are dense enough, you are not simplifying the original data. In some cases and depending on the data, the differences may be small enough that they don't matter, but I guess it depends on how fine of a deviation or difference you are looking for.

Eugene
WWilford
V.I.P Member
V.I.P Member
Posts: 295
Joined: Mon Jul 01, 2013 12:39 pm
10
Full Name: Warren Wilford
Company Details: Berkshire Dimensions
Company Position Title: owner
Country: USA
Linkedin Profile: Yes
Been thanked: 17 times

Re: Deformation analysis statistics

Post by WWilford »

I would suggest scanning a known flat surface such as a granite surface plate. A local machine shop may have a large one that you could scan.

Warren
StuEv
I have made 50-60 posts
I have made 50-60 posts
Posts: 52
Joined: Mon Nov 16, 2015 12:52 am
8
Full Name: Julian
Company Details: University
Company Position Title: Student
Country: Germany
Linkedin Profile: No

Re: Deformation analysis statistics

Post by StuEv »

3DForensics wrote:...for the reasons stated above, I was suggesting to use two dense point clouds. As pointed out, there can be problems either way, but if the point clouds are dense enough, you are not simplifying the original data. In some cases and depending on the data, the differences may be small enough that they don't matter, but I guess it depends on how fine of a deviation or difference you are looking for.

Eugene
Well is 1cm mean spacing enough for this kind of analysis? i could unify the point cloud again with less spacing so i have a more dense point cloud. It is all about the triangle size i think, if i use dense point cloud for a mesh, it won't fit between the noise because it takes all the points to create the mesh, but i should get besser results wenn comparing it with a point cloud. i am looking for a milimeter deviation because the range accuracy of hds has also 1mm accuracy.

@3DForensics:

well i do want to use the mesh as the reference object to compare my point cloud data. to investigate the accuracy of the scanner with the wall i used, is pretty hard because it is a dry wall without mortar between the fugues and it is also not flat.
The problem is, my project manager wants me to investigate the accuracy of the scanner on this wall anyhow...

how many iterations of refining the mesh with points of the point cloud would you suggest? to get accurate results, should i set the point deviation to zero?

and last but not least: if i make the comparison between two epoches to investigate the accuracy of HDS7000, i have discuss, on how many milimeters i can expect deformations with the hds7000.

What influences are involved to get to the result?

vertex accuracy(?) +
registration results(2mm mean error)
Linearity Error of HDS <1mm+
horiz. and vertical angle accuracy 0.007°+
results on the mesh to cloud comparison = +-2

its difficult to for me to discuss :cry:

Thank you for your help
User avatar
danielgm
V.I.P Member
V.I.P Member
Posts: 195
Joined: Thu Jun 20, 2013 10:45 am
10
Full Name: Daniel Girardeau-Montaut
Company Details: CloudCompare
Company Position Title: Administrator
Country: France
Linkedin Profile: Yes
Location: Grenoble, France
Has thanked: 11 times
Been thanked: 34 times
Contact:

Re: Deformation analysis statistics

Post by danielgm »

Didn't Gilles said that 'the mesh is really fitted in the middle of the point cloud thickness'. Shouldn't it ignores the noise then? Because if you subsample (or resample?) the cloud then you have to wonder how this resampling will affect your measurements. Depending on how it is done you might lose some crucial information.

And if I'm not mistaken, instead of the scanner accuracy you are trying to determine the sensitivity of your measurement process, i.e. what is the smallest displacement that you can detect with your whole process (meshing steps included).

If yes then you should definitely take a look at the M3C2 distance computation algorithm (http://arxiv.org/ftp/arxiv/papers/1302/1302.1183.pdf - there's a plugin in CloudCompare to compute this distance ;) ):
  • It's the most robust way I know to compare two unknown surfaces without any intermediary process
  • it takes the measurement noise into account (i.e. you can use the original raw point cloud)
  • it generates a confidence value associated to each measurement
m3c2_principle.jpg
The authors had the exact same goal as yours: determining the smallest change as possible with a standard TLS on unknown (and complex) shapes.
You do not have the required permissions to view the files attached to this post.
Daniel
CloudCompare admin
http://www.cloudcompare.org/
StuEv
I have made 50-60 posts
I have made 50-60 posts
Posts: 52
Joined: Mon Nov 16, 2015 12:52 am
8
Full Name: Julian
Company Details: University
Company Position Title: Student
Country: Germany
Linkedin Profile: No

Re: Deformation analysis statistics

Post by StuEv »

Well, ur right i want to investigate the sensitivity of my measurements of the two epoches, but as a result i can also tell how accurate scans are with the target configuration is made.
After registration i got a mean error of 2mm but my configuration is pretty bad because i was only to place the targets in the same direction, so the z axis is still flexible.
It seems that you cannot use the corrections of the dualaxis-inclinometer in cyclone to level your scan accurate enough (you can only use the corrections with a filter in Z+F Control, prette awkward).
without the usage of the dual axis inclinometer i am a victim of the bad configuration... i now compared the two epoches of HDS 7000 and you can see the rotation in x direction between the two epoches:

For meshing i did it with 1cm mean space and 10 cm hole detection and then i refined and smoothened the mesh.
this is the result:
HDSE1E2Reshap.png
as you can see it gous from -2mm to 4mm from bottom-up, thats because the scan is not accurately leveled becuase only an inclinometer is built in :?

Thank you for your advice yet :-) creat article daniel!

StuEv
You do not have the required permissions to view the files attached to this post.
Post Reply

Return to “Leica Cyclone 3DR”