CloudCompare forum

Hello Mr. Daniel,

I sincerely appreciate your ongoing support on this platform.

Currently, I'm immersed in my research project where I aim to compare two meshes, both exhibiting a cylindrical shape. During my exploration, while examining the C2M signed distances for one sample, I noticed that the points at the lowest section of the control mesh hover around 0.6 mm (despite the distance between these points in the control and reference meshes in this bottom area being approximately 3mm - figures 1 and 2). Moreover, in another comparison, the points register a value closer to 0.7 units (although the distance between corresponding points in the control and reference meshes at the bottom area is around 5mm - figures 3 and 4).

I recall your earlier explanation that C2M operates by measuring the distances between the closest points. Thus, in scenarios like this, the points at the base of the control mesh are juxtaposed with points from the reference mesh situated in the periphery at a similar level, rather than directly with the lowest points of the reference mesh. Am I understanding this correctly?

Essentially, what I aim to achieve through this comparison is to ascertain the deviation across all meshes and to ensure that I'm comparing points located at equivalent positions in both the control and reference meshes. Could you provide guidance on how I can accomplish this?

My ultimate goal is to document how altering the depth of the scanned object impacts the accuracy of the scanner. I suspect that relying solely on the RMS generated through the C2M distances might underestimate the overall discrepancy, what do you think?

Can you send me these meshes so that I can double check things on my side? (the new version of CC is more robust, so the C2M distances should not have the same issue as before with 'flipped' distances, but I don't see why the reported distances would be much larger than what you expect...).

Essentially, what I aim to achieve through this comparison is to ascertain the deviation across all meshes and to ensure that I'm comparing points located at equivalent positions in both the control and reference meshes. Could you provide guidance on how I can accomplish this?

Can you clarify what you are referring to by 'equivalent' positions? Because the issue is that neither the vertices of the mesh, nor the points sampled on the mesh will be exactly at the same position. You are kind of bound to use a statistical approach. I believe that using a sampling of points with a controlled surface density (the second option when using 'Edit > Mesh > Sample points') will be the best option in this case. I mean sampling points on the compared entity with a controlled density, and comparing the resulting cloud to the reference mesh.

My ultimate goal is to document how altering the depth of the scanned object impacts the accuracy of the scanner. I suspect that relying solely on the RMS generated through the C2M distances might underestimate the overall discrepancy, what do you think?

You can of course compare the average distance if you want to detect a potential shift. But that depends on how the different datasets are registered together (if it's with ICP or a similar approach, the average distance should always be close to zero, as these techniques will suppress the shift/bias).

And adding the maximum distance can also be interesting, but sadly it's highly dependent on the noise... So not super robust.

In the end the RMS, or equivalently the standard deviation, are probably the best indicator AFAIK.

Hello again, Mr. Daniel. I've emailed you (daniel.girardeau@gmail.com
) with the reference and control meshes attached.

My main concern revolves around scanning teeth with holes of varying geometries (differing depths and widths) to assess the digital scanner's accuracy against reality (Micro-computed tomography scan). As the depth of the hole increases, the digital scanner struggles to illuminate the base of the hole adequately. For instance, in the provided meshes, the scanner only detected around 4.9 mm out of 10.5 mm, a substantial deviation, especially concerning the fabrication of a post for tooth insertion.
The issue with employing RMS in this study stems from the fact that as the depth of the hole increases, the reference scan expands, while the control depth remains relatively similar in depth due to the scanner's light limitations. Consequently, despite varying hole depths, the RMS values remain closely clustered. Relying solely on RMS would thus fail to highlight significant differences.
I am exploring a method to quantify the overall difference between the two meshes, whether for example by measuring the volume or the surface area of the entire meshes. This is because C2M distance measurement alone won't account for the absent portions (see the selected part in the image) not captured by the digital scanner( cloud mesh).

Hi,

I finally found the time to look at your data.

First, that's probably a detail, but since the radius of the hole is around 0.75 units, I think the distance of 0.7 units for the central part of the bottom of the hole is not inconsistent: But anyway, you are correct, the RMS of the distances, or even the distances alone won't help you detect how large is the missing part. Unless you compute the distances in both ways (Reference-Control and Control-Reference) and then take the largest value.

Computing the volume is also a good option. If you want to use CC for that, you'll have to make sure the holes are properly aligned with an axis (e.g. Z), then sample a lot of points on the 2 meshes, and eventually use the 2.5D Volume computation tool.

Thank you once again, Mr. Daniel, for your response and continued support.

I have one more question if you don't mind.

The scanner I am evaluating attempts to autofill areas when it cannot capture the bottom of the actual hole, resulting in noticeable discrepancies at the bottom of the model, with the largest being 0.7. This issue is evident in the model I shared with you.

This autofill does not accurately represent the actual area in the reference model, where only a peripheral wall is present.

Is there a way in the software to automatically exclude distance measurements of points in the control mesh/cloud that don't have corresponding points in the reference model, similar to this example? While I can do this manually through segmentation toll, it is prone to human error and would not be consistent across all samples.

Not with the C2M tool... M3C2 has a concept of 'direction' and max distance to the origin point, but you would have to sample points on the mesh first to use it. And the number of parameters is quite higher...