Volumes are generated using survey shoot and LRDUs. To improve the cave representation, I temporarily define the splays as survey shots. It gives the cave a bit of a hedgehog shape but this improves the Volume representation in my opinion. It would be great to have the splays represented in the Volume.
I wasn't thinking about something that hard :). Maybe a first step could be to get the splays as tube (1m diameter) from the station sticking out of the volume... like a hedgedog.
Volume:
Volume using splays
The glitches do so (at the moment) is that LRUD's changes due to the splays becoming survey shots... but if displaying the splays had no effect on the LRUD's, then it would be better.
This said, I like the idea of triangles joining the splays.
Maybe you could have a look at Cloud Compare. This is an open source software for handling/processing point clouds (I use it for my Caveatron surveys). There are features in there using Delaunay triangulation. Also Marco Corvi, developer of Topodroid, has developed a secondary software called Cave3D using 3 differents methods to compute cave walls.
Hi Jochen, I’ve been playing with Cloud Compare to create meshes from point clouds (a lot more points/vectors than with splays), and I get a far better results for cave modeling using the plugin « Poisson Surface Reconstruction » than with Delaunay based tools. Worth having a look.
Would Claude helps getting the best 3D walls ever using splays? I'm still using Therion for 3d volume as it creates a "volume" based on splays. It's fast and the viewing is very flowing. But this means setting up Therion, and it's a pain.
This is an example of MeshLab. Please send me a small cave, so I could convert it to .ply
Here is the complete workflow in MeshLab to generate a mesh from your point cloud:
Import
File → Import Mesh… → select the .ply file. Initially you will only see points. If nothing is visible: activate the "Draw Points" icon in the toolbar or zoom in using the scroll wheel / Ctrl+H (Reset View).
Check normals (optional, recommended)
Render → Show Normal/Curvature → displays the normals as lines, so you can see whether they actually point outward. Length too large/small? Adjust under "Normal Length" in the same menu.
Reconstruction Depth: start with 8. Higher (9–10) = more detail, but also more noise/computation time. With only 174 points, 7–8 is usually sufficient. Pre-Clean: check the box.
Click Apply. A new layer (the mesh) will be created.
Poisson generates a closed surface and "fills in" large gaps — with only 174 survey points the result will be very smooth and bulgy. This is normal with such sparse point density.
Trim excess geometry
Poisson often generates a bubble that extends beyond the actual data:
I've been experimenting with various software tools for the past two years, trying to find a satisfactory solution for producing cave wall models from survey data. CloudCompare, MeshLab, and many others — either alone or in combination — have all failed to deliver a convincing 3D volume. As you pointed out, point cloud density is a core issue, and even with a high splay density, standard point cloud tools simply don't cut it. So far, no software has proven as effective as Therion — and I hate to admit it, because it is arguably the least ergonomic, least user-friendly tool imaginable. Yet it remains the most efficient at producing pseudo-3D models. The fundamental problem is that virtually all available tools are designed for generic point clouds and ignore one basic but critical golden rule of cave surveying: a wall can never be crossed by a splay or a station-to-station shot. Therion works around this with a trick that quickly generates a lightweight pseudo-3D volume, easy to manipulate in a 3D viewer — but it's still not a proper wall reconstruction. I've had a few attempts with Claude, starting from a CRP-produced .dxf file, and so far the results using Poisson reconstruction, Delaunay 2.5D meshing, or a Signed Distance Field approach enforcing the golden rule have been far more promising than anything I've achieved with conventional software. That's exactly why I like the idea of you, Jochen, challenging the current assumptions — because if the golden rule is properly baked into the reconstruction algorithm from the start, rather than treated as an afterthought, we might finally get somewhere worth going.
Therion uses its paroibuilder3D algorithm, which works as follows: 1. Grouping by station: the splays are grouped by measurement station and sorted by azimuth.
2. Cross-sections: cross-sectional profiles are constructed perpendicular to each centreline segment, by interpolating the splays’ impacts on this plane.
3. Inter-section connection: adjacent profiles are triangulated (local Delaunay) to form gallery tubes.
4. export to mesh format
It does requires a bit of splays cleaning work (something the mesh produced by long splays is bad a requires to disable some off them) I find the visual result generaly very convincing.
Is this algorithm open-source and could be used as a starter? A starter because the mesh produced isn't clean at all. so it would requires some more step:
a - identify all triangles intersection
b - cut triangles to match intersection
c - remove all inner triangles (to keep the walls and only the walls.
Also, paroibuilder3D uses LRDU in the model, which is bad when you have enough splays... LRDU are poor quality measurement and are often calculated from splays. if there are splays... the model shouldn't use LRDU I guess.
What I like is that the cross-section planes from CaveRenderPro are used. If splays are available, they are used for the profile; otherwise the LRUD values are used, and if those are absent, the default passage dimensions.
CaveRenderPro
Thema drucken
10.03.2021 12:05
Antworten
