Integrate surface reconstruction approach(es)

[puzzlepaint]

It is often desirable to obtain a surface (mesh) reconstruction rather than only a point cloud, which is the current reconstruction output of BAD SLAM. Some possible options might be:

• Integrate Poisson Surface Reconstruction as a post-processing step (such that it can be conveniently invoked with a single click from the GUI)

• Integrate SurfelMeshing for live meshing during the reconstruction. A good integration might not be straightforward however, since:

  • BAD SLAM is constantly creating new surfels and merging surfels in existing areas, which SurfelMeshing might have trouble keeping up with.
  • For SLAM, one wants to sparsify the surfels a bit to keep it fast, while for reconstruction, the surfels should be as dense as possible.

[kirbiyik]

Is there any update on this feature @puzzlepaint ?

[puzzlepaint]

I created the issues with the "enhancement" label to show what would be useful additions in my opinion, but this does not mean that I have any plans to work on these. So, no updates should be expected. If anyone wants to contribute a PR with those features, it would however be welcome of course.

Regarding this particular issue, note that it is already possible to create a densified point cloud, save it to disk, and run something like Poisson Surface Reconstruction on it manually to get a mesh reconstruction.