Double counting due to tile overlap

[m-scholes-99]

Hi Thomas and all,

I have been using raycloudtools to extract volume estimates from my TLS data. To do this, I have tiled my data and introduced and overlap, currently set to 5m. In these areas, I am finding that some structures are getting segmented to multiple trees, depending on which tile it is in. This means, some structures are being reconstructed multiple times, which is causing issues for my volume estimates from the plot.

Initially, I thought this was just for full trees, and I filtered out these cases by only selecting one tree in instances where there were multiple trunks bases in the same location. However, it is not only these cases, but also situations where branches are assigned to 1 tree in 1 tile and to another tree in the second. Therefore, there is no metadata to signify that the points are assigned to multiple structures so they are difficult to filter.

I was wondering if anyone had come across this issue and had thoughts for solving it? My initial ideas revolve around finding a common volume in the resultant tree mesh structures, but this will be quite complex to implement and require a sort of arbitrary threshold to determine what overlap is considered significant. I thought it would be best to ask if any solutions had been integrated into this toolkit that I had missed, to avoid "reinventing the wheel".

Any advice from developers or users would be very much appreciated!

All the best,

Matthew

[thomas-lowe]

Hi Matthew, I'd like to help you resolve this, but I'm having difficulty picturing the problem. If you are able to send test data or a screen shot it might help.

Just to check we're on the same page:
to use rayextract trees on large datasets you can use say: raysplit 50,50,0 5 which includes a 5 metre overlap on the grid cells.
Then after running rayextract terrain on each cell, you can call: rayextract trees cell_x_y.ply cell_x_y_mesh.ply --grid_width 50
This --grid_width option tells the rayextract to cut out all trees with root point outside the 50m cell. So it has overlap to generate the tree, but removes the overlap on saving out the tree file.
Consequently, each cell_x_y_trees.txt should contain unique trees without any doubling up.

If you have done this and all trees are less than 5m in crown radius then it sounds like a bug and I'd probably need to see pictures or get some data in order to figure out what is happening.

Happy to help.

[m-scholes-99]

Hi Thomas,

Thank you so much for such a prompt response!

Yes, I think you understand the problem but I have attached some images just to be sure:

On the top image, there is this pink branch, reaching off to the top right, and the same branch, on the bottom image, is assigned to a different (cyan) tree.

Due to the format of the data I had to tile the data before using any raycloud commands so that it could be in an accepted format. But I think using the --grid_width command should be able to help with this issue.

Just to be sure I understand; if my tiles are 10x10m with a 5m overlap and I run rayextract trees tile_x_y.ply --grid_width 10 then this will use all points within the tile, and the boundary. However, it will only segment trees that have a trunk base within the tile, and not include those within the boundary?

If so, I should be able to reprocess the data with this and hopefully that should fix the issue.

Thanks again for the response!

Matthew

[thomas-lowe]

That's correct Matthew. To be clear here's a diagram for your example:

Grid cells are centred at multiples of 10m x 10m, so if you aren't using raysplit grid then you need your tiling to do the same. If you can't retile in that fashion then you could just translate before running rayextract trees (e.g. raytranslate cell_x_y.ply -5,-5,0) then translate the results back (treetranslate cell_x_y_trees.txt 5,5,0).

And the overlap is 5m, so each enlarged cell is only expanded by 2.5 m. So the crown radii should be under 2.5 m.