Rayextract trees QSM branching off too early

[rvos94]

Hello Thomas and others,

I am using raycloudtools on a hedge row of trees that is pruned regularly. My scans were taken with a TLS. I ran into some interesting behaviour with the QSM where it branches off halfway the tree stem.

Input used:
rayextract trees 'file_raycloud.ply' 'file_raycloud_mesh.ply' --crop_length 0.01 --global_taper 0 --global_taper_factor 0 --gravity_factor 0.5

As can be seen in the pictures below, the tree branches off halfway of the stem. From this point, a smaller branch than the original stem size is going up alongside the stem with very thin side branches coming out to 'collect' the remaining points of the stem.

Point cloud

Resulting mesh

Resulting mesh tree trunk

I was wondering if this can be fixed on my end by altering my input command or if this a model limitation?

Thanks for working on this amazing tool and kind regards,
Ruben

[thomas-lowe]

Hi Ruben,
it looks like the stem is splitting too early.
As the algorithm traces up through the stem points it clusters the points based on a gap width, so if there is a big gap between points on the stem it will think it has become two branches.
The gap width is specified as a ratio relative to the branch length to tip, and is an internal constant that defaults to 0.016. You can change it with the -g (--gap_ratio) parameter. I'd try doubling it and see if it helps.
You could also try upping the span ratio (--span_ratio) from 4.5, which makes it less inclined to split, but probably won't do much.

I think there is a bit of a model limitation also. You hedge has a fairly thick stem right to the top because it has been cut. This breaks the similarity approximation which sets branch radius in proportion to branch length at each bifurcation. So it "wants" your stem to be getting smaller than it is.

Radius estimation is the biggest difficulty in tree reconstruction, so it needs to make some approximations unfortunately.

[rvos94]

I have tested multiple values for both parameters and sadly both do not seem to do much to prevent the early splitting.

[thomas-lowe]

Yeah OK, it must be the second reason (similarity approximation). Generally works well on natural trees but can be a problem on pruned trees where branch diameter and branch length are uncorrelated.
Is it possible to get that point cloud? It would make a good test case for improvements to the radius estimation.