[OSM-dev] Fwd: [GSoC] A GPX photo stamper.
lists at humanleg.org.uk
Fri Apr 3 12:53:00 BST 2009
You should probably take note of the NL / OpenStreetPhoto guys' efforts to do
SIFT on streetsigns. It isn't as simple as you might think.
One thing I will say though is you might get better results looking at the
Affine SIFT (ASIFT) algorithm instead.
On Friday 03 April 2009, Tijs Zwinkels wrote:
> So, I've decided to take on this project. :)
> I'm currently writing my proposal, which I'll post to the wiki and submit
> at the GSoC site in 1 ~ 2 hrs time. (let me know if you want to see the
> intermediate versions, I can just post to the wiki whatever I have already)
> Just bouncing a few last ideas, mainly to define the scope of the project:
> - I've decided to focus on the automatic reading of street signs, and not
> specificly on the recognition of other items in the environment. However,
> one of the most challenging parts of this project is to segment
> street-signs from images succesfully, and I'll be using SIFT (among other
> things) for that. I fully expect the toolset that I develop to be very
> useful in recognizing other objects as well.
> - Since street-signs look differently in different parts of the world, for
> now I'll just try to automatically read dutch street-signs. I think this
> will give me enough variance to struggle with for now. :) - ofcourse, it
> shouldn't be hard to adapt this work to foreign street signs.
> - There is a large difference between whether the software should be able
> to catch and read street signs that just happen to be in a picture, or
> whether people took a picture with the intention of capturing a street
> sign. (by which we would view the sign from a relatively straight angle
> with a reasonably high resolution, not too much reflection, etc)
> For now I'll assume the latter case. If the performance is reasonable, we
> can always scale up to the more difficult cases.
> - Tijs Zwinkels
More information about the dev