RANSAC SD crash

[sguinard]

Hello,

I'm new with CloudCompare, and wanted to try RANSAC plugin in order to detect trunks in a point cloud.
The problem is that I can set the parameters but it seems that the software freezes just after that.

Could it be linked with the size of the point cloud ? Or is there an other cause ?

[daniel]

Does the software actually crash or it just freezes?

And what is the size of the cloud? (because yes, it can take a while, especially if the plugin needs to compute the normals first).

[sguinard]

Thank you for the reply,

Actually it just freezes

My point cloud contained 5 000 000 points but when I try with a 1 000 000 one it works in 30 seconds so I guess the size is the main cause ...

[daniel]

Yes the code used by the plugin is the original one (from Schnabel's library: http://cg.cs.uni-bonn.de/en/publication ... efficient/). Therefore it's a research work (even though it's already quite good and no real bug has been found yet).

[sguinard]

Thank you !! I just have an other question :

is it possible to add a filter or something like that, in order to select only the most interesting results ?

Best regards

[daniel]

Not really. You can only select the type of entity.

And it would depend on what you call 'interesting' anyway ;)

[sguinard]

What I was thinking about is, in my case, keep only cylinders that have a certain radius.

If it's not possible, do you think I'll have to integrate my own functions to RANSAC plugin ?

[daniel]

Yes, I think it would be the best way. And you get something working (even in a dirty way) we can integrate it afterwards.

[sguinard]

Hi Daniel,

I think I'll work on it soon,
I just also need to know if it is possible, once you obtained 3d primitives via RANSAC plugin, to segment your point cloud using those primitives ?

[daniel]

Well normally the plugin output the subset of points corresponding to each primitive

But you could also compute the distances between the whole cloud and the primitive to segment the points below a given threshold distance.