I have a very dense point cloud (billions of points) of the exterior of a building obtained by laser scanning it with a Leica head. I successfully subsampled it down to around 500,000 and I'm trying to print the building by first creating a mesh. I tried using CloudCompare, Meshlab and PDAL, using Poisson surface reconstruction. However, the resulting mesh is full of holes, mainly in the roofs which have the lowest point density, and I cannot print it. Is there any algorithm which could use the fact that the point cloud is precisely the exterior part of a geometric thing?
Yes there are similar algorithms, but (afaik) not as ready to use programms. I wrote a bachelor thesis by my own, where i converted poind cloud datas of scanned surfaces into contour octrees. This based on the work of Laine (https://users.aalto.fi/~laines9/publications/laine2010i3d_paper.pdf) and the approach of using sparse voxel contour octrees, but instead of using polygons it used point clouds. This way was intended to get fast, good approximated results for visualizing.
But there may be also other slower and more accurate algorithms.
Btw. this question is not good placed in the 3D printing forum, because it is a question about data conversion.
Filling out holes in a mesh created using large liar data points is a mess.
The actual way to do it in the industry is to manually fill all holes ( Yes I know it takes for ever to do and people get paid to do this )
Import the mesh into Autodesk 3Ds max and go about fixing the holes one at a time if you want it to be accurate or select all and use the cap function .
Hope this solves your problem .