Sunday, February 17, 2013

Dayton Masonic Temple Photogrammetry

Bundler-PMVS2 Dense Point Cloud Visualization in Meshlab
I wanted to experiment with some free photogrammetry software, and the Masonic Temple in Dayton is a nice target of opportunity. There are some free (as in beer) options, but I wanted something that was free (as in freedom) that I could run on my own machines rather than in a software-as-a-service cloud. The basic work-flow is demonstrated in this post by Andrew Hazelden for some aerial photographs. He gets fairly impressive results using only free software.

The basic steps to turn a bunch of picture into a 3D model are
  1. Run Bundler which estimates camera positions based on matching keypoints between pictures.
    I had to make some minor tweaks that are not documented on the main sites, but are out on the interwwebs: slight source code modification, add executable permissions
  2. Run CVMS (in the case of a large number of pictures) which clusters views so that the dense reconstruction can be accomplished efficiently in parallel.
  3. Run PVMS-2
  4. Import the ply file created by pvms2 into meshlab
  5. Calculate normals for the point cloud and run Poisson surface reconstruction; transfer point colors to the new mesh.
  6. To Do: get the latest meshlab and try raster layer projection onto the mesh
I need more pictures with more overlapping to get better results; but this is promising.




Update: Now I need a little quad-copter to constrain things from angles I can't get from the ground.

3 comments:

  1. Updated to include Poisson surface reconstruction with additional pictures around the sides and of the front facade.

    ReplyDelete
  2. This basic tool chain demo'd in an extensive post by Jessie Spielman by way of Make.

    ReplyDelete