Question

Unwanted result using the pointcloudsurfacebuilder with poisson construction mode

  • 31 August 2018
  • 8 replies
  • 6 views

Badge

Hi guys,

I have this point cloud that I have filtered and clipped so that only vegetation areas remain. These areas I want to build a mesh of so that I can import this as a 3D-model into another software. I use the pointcloudsurfacebuilder for this purpose. Using construction mode "simple triangulation", everything works fine, a separate mesh is created for eache area of vegetation.This mesh is quite coarse so I want to use constrouction mode "poisson" instead. This creates a much smoother mesh, only problem is that it also creates a rectancular surface around the whole extents of my pointcloud thus making the mesh useless for me.

Anyone here who can help me sort this issue? See attached screenshots of both results.

 

 


8 replies

Badge

Still struggling with this issue, any suggestions could be of great help to me. :)

Badge +11

Hi @atle_hoidalen,

I'm not 100% sure on this one, but perhaps you could create the boundary features of your 'simple triangulation' vegetation mesh, following this tutorial. Then with those closed polygons, you could clip the poisson construction output to those vegetation boundaries. This article on How to clip TIN Surfaces, might also be a good resource, though, since it uses the SurfaceModeller, perhaps you will just want the Clipper. If that doesn't work out, would you mind sharing a sample of your data? If you don't wish to share your data to the public, you are welcome to submit a case with us here.
Badge

Hi @atle_hoidalen,

I'm not 100% sure on this one, but perhaps you could create the boundary features of your 'simple triangulation' vegetation mesh, following this tutorial. Then with those closed polygons, you could clip the poisson construction output to those vegetation boundaries. This article on How to clip TIN Surfaces, might also be a good resource, though, since it uses the SurfaceModeller, perhaps you will just want the Clipper. If that doesn't work out, would you mind sharing a sample of your data? If you don't wish to share your data to the public, you are welcome to submit a case with us here.
Hi!

 

Thanks for you suggested workflow. I have tested it and though it might produce the desired result it does create a quite complex workspace. I would rather learn if it is possible to tweak the poisson triangulation to produce the desired result directly. I will test some more and maybe upload my workspace here. :)

 

 

Badge

Still no answer from Safe on this one. I would very much like to see a walkthrough of this transformer and especially using the poisson tringulation.

Badge +11

Still no answer from Safe on this one. I would very much like to see a walkthrough of this transformer and especially using the poisson tringulation.

Hi @atle_hoidalen,

sorry to hear that you're still searching for an answer with the poisson triangulation. Would it be possible to share a sample of data? I'd be happy to look into it further with some data to test on.

I took a look into some previous uses of the transformer, and in one case it looks like the Simple Triangulation is recommended over the Poisson as the results tend to resemble the input data far more than the Poisson. For the rough output that you're getting, it looks like you can use the parameters to limit the angle between adjoining surfaces in order to limit the noise in the output. Without looking at your data, I can't say for certain what those parameters should be though.

Badge

Hi @atle_hoidalen,

sorry to hear that you're still searching for an answer with the poisson triangulation. Would it be possible to share a sample of data? I'd be happy to look into it further with some data to test on.

I took a look into some previous uses of the transformer, and in one case it looks like the Simple Triangulation is recommended over the Poisson as the results tend to resemble the input data far more than the Poisson. For the rough output that you're getting, it looks like you can use the parameters to limit the angle between adjoining surfaces in order to limit the noise in the output. Without looking at your data, I can't say for certain what those parameters should be though.

Hi @JovitaAtSafe. I' uploading a simple workspace and an example las-file that I've been testing on. My goal is to create a closed smooth mesh from the point cloud for use in visualization purposes. First two examples are done with the "simple triangulation" method, example 3 and 4 with the poisson method. Poisson is what should give the result I want, but as you can see, it seems to create some weird surfaces at the edges in my model and also, look at the highes building, it creates a ballon-like roof even when there is no points as high as the triangulated surfaces are. By reducing the scale factor in the poisson triangulation (example 4) i get slightly better results but no where near what you can get in other software.

Zip-file with examples uploaded in my initial post.

I hope you can bring in someone that knows this transformer and guide me through the process to get the results i want.

Best regards, Atle

Badge +11

Hi @JovitaAtSafe. I' uploading a simple workspace and an example las-file that I've been testing on. My goal is to create a closed smooth mesh from the point cloud for use in visualization purposes. First two examples are done with the "simple triangulation" method, example 3 and 4 with the poisson method. Poisson is what should give the result I want, but as you can see, it seems to create some weird surfaces at the edges in my model and also, look at the highes building, it creates a ballon-like roof even when there is no points as high as the triangulated surfaces are. By reducing the scale factor in the poisson triangulation (example 4) i get slightly better results but no where near what you can get in other software.

Zip-file with examples uploaded in my initial post.

I hope you can bring in someone that knows this transformer and guide me through the process to get the results i want.

Best regards, Atle

Hi @atle_hoidalen,

Thanks for sharing your workspace and data! I checked in with our development team on the transformer and found out that the Poisson method of surface recalculation is meant to create a single water tight surface.

It is best used for creating a single mesh representing something organic, such as a face or anything in nature, and not meant for artificial structures like buildings due to the way it calculates and smooths the mesh, which would explain the bulge in your building. Since the Poisson method creates the single mesh, this explains why your output includes the full extent rather than just giving you the vegetation patches.

Long story short, unfortunately while the Poisson triangulation seems to be creating a more desirable output visually, the algorithm may not be the right fit for what you are trying to do. If you are still wanting to use the Poisson triangulation for the vegetation, something that you could try is to split each area of vegetation out into it's own point cloud to write out into individual meshes, though that might involve creating boundary polygons for clipping. Just an idea though (:

Lastly this was another article that I was looking at as a potential solution. Out of curiousity, can you tell me anything more about the bigger picture of this project? What kind of format or software were you hoping to write to? Perhaps I can suggest another format or method that may be a better fit?

- Jovita

Badge

Hi @atle_hoidalen,

Thanks for sharing your workspace and data! I checked in with our development team on the transformer and found out that the Poisson method of surface recalculation is meant to create a single water tight surface.

It is best used for creating a single mesh representing something organic, such as a face or anything in nature, and not meant for artificial structures like buildings due to the way it calculates and smooths the mesh, which would explain the bulge in your building. Since the Poisson method creates the single mesh, this explains why your output includes the full extent rather than just giving you the vegetation patches.

Long story short, unfortunately while the Poisson triangulation seems to be creating a more desirable output visually, the algorithm may not be the right fit for what you are trying to do. If you are still wanting to use the Poisson triangulation for the vegetation, something that you could try is to split each area of vegetation out into it's own point cloud to write out into individual meshes, though that might involve creating boundary polygons for clipping. Just an idea though (:

Lastly this was another article that I was looking at as a potential solution. Out of curiousity, can you tell me anything more about the bigger picture of this project? What kind of format or software were you hoping to write to? Perhaps I can suggest another format or method that may be a better fit?

- Jovita

Thanks @JovitaAtSafe!

My intensions for doing this is that we want to use the point cloud for creating a "Google Earth-like" mesh-model for visualization purposes in our projects.

Reply