M3C2 Parameter Questions

[Geononymous]

Hi everyone!

I am a masters student in the geosciences and i want to do mass change detection for an active dune system (as high resolution as possible). I got 2 surveys, each referenced with RTK dGPS with overall GCP quality of 4-6cm RMSE. The dense point cloud was computed with Agisoft and exported as *.PLY containing normals from agisoft. Ground Sampling distance is 1.4cm/pix. The normals were visually inspected in Cloudcompare and seem to represent the surface roughness ideally. The area is about 550x500m and each cloud about 500mio points. Ridge of the dune is 30m high.

My question now: What parameters should be set into the projection field "diameter" and "max depth" and what do they do. When i check "use normals from cloud one" and click "guess params" i get 0.08 for diameter, and for max depth i get 45.88. And how do i obtain registration error? I read you have to align the 2 clouds somehow. Isn't that for georeferenced clouds already the case? Or do i have to have a fix point inside the clouds and check them against this fix point somehow?

Greetings

Geon

[daniel]

It seems that you have very nice datasets ;)

Have you read the M3C2 wiki and article? (http://www.cloudcompare.org/doc/wiki/in ... 2_(plugin)). You should indeed definitely use the cloud's normals. The 'depth' is the maximum search distance for equivalent points in both clouds (to avoid losing too much time searching for very far points or potentially wrong correspondances). You can ignore the 'Normals' diameter (as you already have them). And the projection diameter is the search cylinder diameter. It depends on how detailed / dense are your cloud and the scale of the 'objects' you are tracking.

You can start by testing the algorithm on a subsampled cloud ('Core points' - this is the default) and once you are happy with the results, you can use all the points.

And for the registration, you can use the GCP RMSE (it's used for the computing the 'confidence': to simplify, all displacements with an amplitude below this value are fishy)

[Geononymous]

Thank you so much for your answer. Yes i am quite happy about the datasets :)

I have read the articles and the literature on how the algorithm works. But it was difficult to understand for me. Thank you for explaining. I can read from your post that a maximum search depth of 45 might be too much and would run a long time to compute. Max depth is a rough estimation on how far the surfaces of the point clouds are away from each other? If that is the case i think 5 meters shoud be sufficient?

Greetings

Geon

[daniel]

Yes it's a rough estimation (and generally bigger than the truth) of how far is the farthest point. It's a good idea to set it yourself if you have some a priori information (just make sure the units are the right ones ;).

[Geononymous]

Hey Daniel

I got results and still have a few questions to it. I set the max depth projection to 5.0 instead of suggested 45.0 / used the normals within the *.ply, and set the registration error to 0.065 because of 6.5cm max GCP RMSE. This took 8 days to finish. Looking at the M3C2 output with the option "show NaN" ticked (1) there are vast grey areas. Is this the case because the "max depth" might have been too low? If it was set to 5.0 how much longer (rough estimation) would it take to compute it with 10.0 or 20.0?

Wiki tells me that for grey points no points from the other cloud could be found in the search cylinder. How come when the option "show NaN" is ticked off (2) there are still points with apparent values where the grey ones were? Can i still use these values to make assumptions?

Same question for the significant change output. Sig. change tells me where more than 0.065m has moved if i am right?

Another question is how to set the color scale correctly. Is there a link maybe with good explanations? Where do i find the value for maximum distance between the clouds? How do i set it for significant change?

For the last question: How can i visualize it for high res figures in my thesis other than snipping tool? And is it possible to load the M3C2 output into Globalmapper? I want to lay the orthophoto above the M3C2 output to make comparisons.

There are so many questions i hope its not too much :)

[daniel]

8 days?! Have you first run on a subset of core points? (this is generally a good idea to avoid losing too much time and realize a parameter is wrong ;).

And yes, probably the maximum distance is too low (but if 5 m. is too big for you, then probably those points are not that interesting?

For the fact that you see points with real values mixed with NaN points is probably due to the high density of your cloud? (how clean are the normals by the way?). It's hard to tell.

And significant change is a little bit more clever than that, but more or less: yes.

You should create your own color scale: https://www.cloudcompare.org/doc/wiki/i ... es_Manager

For high res images: https://www.cloudcompare.org/doc/wiki/i ... er_to_File

[Geononymous]

Alright thanks for your help again. Subsampling is a good idea, tried it and it worked. Created my own color scale and i could import polylines with elevation information into CC. So far so good.

Now for presentation of figures i would still need a bit help. The render to file option is good, but i dont understand it well.
When looking at the whole project (1) i get good resolutions without empty lines in between (are those lines faster render features?) but as soon as i want to get into detail, look at a specific part, i get these lines (2) (3), even though enough points should be there to fill the lines. Making (1) as big as possible doesnt show interesting parts in fine resolution unfortunately. Is there a way? How do other people generally visualize their M3C2 results?

A big thanks until now for your help.

[daniel]

You simply have to make the points bigger! (use the global interactors that appear when the mouse hovers the top-left part of the 3D window).

You may also have to let CC 'auto-scale' the points when you render the image (if you use super-resolution). See the corresponding checkbox in the tool dialog. If it's not sufficient, again you can increase the points size.