Photogrammetry: update

Posted by vvoovv on 10 March 2010 in English (English)

In order to simplify installation and further development we have rewritten Bundler's script in Python. Details about their installation and running as well as best practices to take photos for Bundler input can be found at the OSM-wiki:

Absolute orientation is unknown for 3D scene reconstructed by osm-bundler. So the problem is how to estimate the up-direction for the reconstructed scene.

One option would be to take into account the observation that "most photos are taken without significant twist (rotation about the camera’s viewing direction), and compute the up vector which minimizes the amount of residual
twist over all cameras." This method is used by the author of Bundler.

Also one can notice that most of buildings have vertical and horizontal edges. One could try to estimate a dominated direction that the reconstructed 3D points form.

If you have other ideas how to calculate up-direction please post them in the OSM-forum:

Login to leave a comment