[OSM-dev] [GSoC] Automatic Street-Sign Detection and Reading - was: [GSoC] A GPX photo stamper.

Tijs Zwinkels openStreetMap at tumblecow.net
Fri Apr 3 15:02:18 BST 2009


I just posted my proposal to the wiki:
http://wiki.openstreetmap.org/wiki/GSoC_Student_Applications_2009#Draft_application:_Automatic_Street-Sign_Detection_and_Reading.
It might still have a few rough edges. Feedback is welcome!

A few specific questions:
- Do you agree that part of the deliverables is to have the street-sign
detector implemented in JOSM? Would this be the right place to implement
this system? Are we even interested in implementing this system somewhere,
or should it just be a proof-of-concept?
- Would anybody be interested in mentoring this interesting project?
Preferably someone with SIFT-experience, but I would be happy to work with
anyone who's enthusiastic about this project. Stefan or Matt maybe? :)

On Fri, Apr 3, 2009 at 1:05 PM, Tom Evans <tom_evans_a at yahoo.co.uk> wrote:

>
> If this could work off a straight-angle photo, but with only a
> mobile phone camera, that could be pretty nice to work with.
>
> I don't really expect any problems with phone-cams. For most computer
vision algorithms, the first step is to scale down the image anyway. Worse
contrast or bad color representation might be a problem though.

On Fri, Apr 3, 2009 at 1:53 PM, Robert Scott <lists at humanleg.org.uk> wrote:
> One thing I will say though is you might get better results looking at the
> Affine SIFT (ASIFT) algorithm instead.
>
> http://www.cmap.polytechnique.fr/~yu/research/ASIFT/demo.html<http://www.cmap.polytechnique.fr/%7Eyu/research/ASIFT/demo.html>
Thanks for the hint! - More robustness to differing angles sounds like
something we could definitely use. I've added the link to my references and
will look into it in more detail.


~ Tijs
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstreetmap.org/pipermail/dev/attachments/20090403/bee13f00/attachment.html>


More information about the dev mailing list