M3C2 plugin - no data on roughness

[ohnoinkimono]

Hi everyone,

Thanks for the great documentation and for providing all the publications relevant to the M3C2 plugin. I worked through it and feel almost confident, that I know what I am doing. However, some interim results of my analyses still confuse me and I hope to shed some light on this matter with your help. ( I am running CloudCompare 2.12.2 on a Windows 10 Pro, 64-bit)

I am comparing two point clouds using M3C2 since I am looking at an erosion feature progressing vertically and horizontally and therefore need detailed information.

1) I generated the point cloud with OpenDroneMap which displayed a massive vertical offset, that I fixed using ICP alignment (with no segmentation beforehand) in CC: final RMS 1.06. You can see the resulting point clouds below, z value displayed.

2018_Z_pc.png (385.57 KiB) Viewed 3573 times

2019_Z_pc.png (437.58 KiB) Viewed 3573 times

2) In the Lague paper it is mentioned that I should set the Normal scale (D) to a value that is at least 25x roughness of the point cloud. The figure below depicts the roughness I calculated on cloud#1. While the overview depicts nothing but grey values (no roughness?!), a zoom in reveals values mostly ranging between 0.02 and 0.3, which would call for a Normals value of 0.5 to 0.75. When I use "guess parameters" the program however repeatedly gives an output of 1.17 normals diameter and 0.59 projection diameter.

2018_roughness_grey.png (37.41 KiB) Viewed 3573 times

2018_roughness_ZoomIn.png (935.89 KiB) Viewed 3573 times

Now my questions:
A) Can this estimation be trusted since the roughness calcualtion I did manually seems a little "off" or would I be better off by extracting smaller samples from the cloud, calculate the roughness and use a mean to derive the normal and projection diameter?

B) When I check calculation mode "vertical" under the "Normals tab", do I only get Nx and Nx as displacement vectors in the resulting M3C2 file?

I really appreciate your support.
Have a good evening,
Soraya

[daniel]

A) Regarding your second point (2), it's obvious that the radius you used to compute the roughness it way too small (grey color means that there was not enough points around each point to compute the roughness). You have to compute the roughness on a bigger radius (until almost all points have a roughness value).

B) Since you seem to have rather 'flat' clouds, you can definitely set the normals to be always vertical. This way you don't need to bother with the normal scale anymore. You'll get displacements along Z only in this case.

C) Don't hesitate to activate the EDL shader to improve the visualization of your clouds (Display > Shaders & filters > EDL)

[ohnoinkimono]

Hi Daniel,

Thank you for your prompt reply.
I followed your suggestions and

A) increased the roughness radius to 1.0 with the following output

2018_roughness_1.0.png (538.56 KiB) Viewed 3530 times

What is remarkable, however, is that the point cloud itself seems to be "layered": when the cloud renders it first presents points with a roughness value of up to 0.09 (blue colours, also see the histogram next to the colour bar). The points "on top" of that layer, however, show values of 0.3 to 0.6 (green to orange). Could my cloud really be that noisy? I wanted the cloud to remain as raw as possible, but maybe it is more helpful to apply a noise filter beforehand? Once again here I have the problem of defining a radius. I guessed the value 1.0, but that is rather unscientific. How can I estimate a good radius value for both calculating the roughness and applying the noise filter?

B) Thank you, that sounds good. I will probably run the horizontal mode as well since I am also interested in the direction of the erosion.

C) enabled EDL shader.

Thank you again!

[daniel]

Well, either it's noisy, or it's due to the laser penetration in the vegetation?

[PennyMolina]

Please tell me, has the reason for this become clear to you?