Hi Geowankers,
I have an interesting problem which I haven't found covered elsewhere
on the Net, and the results would be generally useful.
SIFT stands for Scale Invariant Feature Transform, and it's used for
picking feature points out of photographs in a similar way to the
human eye. After shoving the results through a Hough Transform, it's
possible to identify and match multiple objects.
It is used in software like Autopano Pro so that photographs can be
taken and then merged together into panoramas of various sizes,
including the complete wraparound situation of 360 degrees with up
and down views. QTVR in other words.
What I'm trying to find now is a way of applying SIFT to 360 degree
equirectangular projection so that I can identify objects inside of
panoramas. Currently SIFT only works on rectilinear projections and
appears to embed that assumption.
SIFT also has licensing issues, so this project may be a better base
to work with:
* Zoran Mesec has written Autopano-Z, a new feature/detection
matching tool that will free hugin from the restrictions of the
patented SIFT algorithm.
Is anyone else interested in doing object identification inside of
equirectangular projection? I am currently extracting a series of
rectilinear photographs for SIFT, but that causes issues when an
object spans a border.
Steve.
--
[EMAIL PROTECTED]
_______________________________________________
Geowanking mailing list
[email protected]
http://lists.burri.to/mailman/listinfo/geowanking
