Photoscan Improvements

Agisoft Photoscan is a new photogrammetry software that finally solves the majority of the problems in photogrammetry.
  • No need to renumber images; the Collada file that Photoscan exports names each camera after the image that it uses
  • No need to downsize images; Photoscan works fine on original resolution DSLR images at 4k + resolutions.
  • Collada export from Photoscan contains all camera position, orientation, and field of view information, along with a reference mesh.
The automatically generated reference mesh is too rough for final use, but it is invaluable as a modeling aid.

Solving with Photoscan

The Photoscan user’s manual provides a good walkthrough of the process. Specific recommended settings include:

  • Use High Quality setting for aligning photos
  • For building geometry, use arbitrary object type, high target quality, sharp geometry type, and the default thresholds.

Splitting Scene into Chunks

For large scenes, sometimes Photoscan will solve a row of images into an incorrect coordinate space — they will match each other, but will not be located correctly into the larger scene.

To solve this, create a new chunk, drag the non-solving images into the new chunk, and solve them on their own. Photoscan provides tools to align chunks automatically, but in practice it’s frequently easier to export the chunks individually, and align them manually in Maya.

Exporting

Export each chunk individually with File->Export Model. Settings:
  • Cameras
  • Split into Chunks
The chunk that is highlighted when the export command is run is the model that is exported.

 

Last Modified: July 22, 2014