hugin-devs team mailing list archive

[Stefan Wirtz <wirtz.stefan@xxxxxxx>]

Dear Hugin Team,

during my time as a PhD student at the University of Koblenz in Germany
I developed a plugin for Hugin (version 0.8), which might be interesting
for you. This plugin allows generating a 3D model of a building
including semantic annotations from image series.

In order to use my plugin, you first need to take overlapping pictures
of the building you want to reconstruct in 3D and then these pictures
need to be undistorted.  Afterwards, you select within Huging the point
correspondences, which are important for the geometry of the building,
from the undistorted image series.

Using structure from motion techniques, my approach is able to recover
the 3D structure and camera poses with these point correspondences. As a
result you receive 3D points, which can be grouped to surfaces and
annotated with labels like wall, window or door. For each annotated
surface the system chooses an image, cuts the surface, calculates the
homography and displays the texture on the surface. After the annotation
phase, we are able to export the semantic annotations and the associated
surfaces along with their texture information to the exchange format
CityGML, an established standard for building  representation.

The approach is accepted at the conference IS&T/SPIE | Electronic
Imaging and will be published in the next months.

The attached document gives an overview of the entire process. So far
the approach only works with Linux.

Other working groups deal with that issue as well. For example:

[1]  http://www.acvt.com.au/research/structure-from-image-sets/videotrace

This approach might be interesting for you, because Hugin delivers a
perfect environment for the issue of generating semantic 3D models.
Therefore, the inclusion of my approach might be an interesting
augmentation for Hugin. I would be pleased, if you see potential in that
approach. I am looking forward hearing from you.


Yours faithfully,

Stefan Wirtz

Attachment: HuginPluginSeMo.pdf
Description: Adobe PDF document