[OSM-talk] GSOC2012 Video Based Speed Limit Detector
kakrueger at gmail.com
Thu Mar 22 01:22:44 GMT 2012
Graham Jones-2 wrote
> Hi Emanuela,
> - I think the project idea is about developing a plugin for the JOSM
> editor? If so it would be good to have a look at that and learn to use
> then have a look at the code and some example plugins to see how they work
> - you should find lots of information on the OSM wiki.
> Hope that helps. Please feel free to ask more questions as you develop
> your proposal.
> I would ask the person who proposed the idea to reply here with a bit more
> information please, as this is not something I know anything about!
I suggested the project on the wiki, and I would also be happy to mentor
this project should a proposal for it be accepted. (Last year OSM only had 3
slots and quite a number of students applied and so it is difficult to say
which ones will get accepted this year).
I'd like to give a bit more background as to the motivation for suggesting
this project and what might be involved in doing it. However, these are just
suggestions and if you have different ideas that would also be great, as
long as you can justify why they would be useful to OSM. In fact,
personally, I would strongly encourage students to enhance the proposals
with some of their own ideas. For one, those are then more likely to
interest the student and ensure that they do a good job with the project,
and secondly it demonstrates that the student has looked into the project
and has a thorough enough understanding of the project to contribute ideas
and to some degree work independently.
In the early days of OSM (and to a good degree still today), people went out
and about specifically to map. They for example walk, cycled, drove along
routes and stopped every few meters to take pictures of street signs, shops
or house numbers or write down notes about what is on the roads, which they
can then easily process back home into detailed maps for OpenStreetMap. This
is really great for the project and often a lot of fun for the mapper, but
also rather time consuming and perhaps not everyones idea of an ideal hobby.
It also becomes less fun if you work in densely mapped areas and are only
verifying data rather than contributing new data. Therefore to scale up
mapping to larger numbers it is important to try and incorporate more
mapping into every day, rather than special mapping trips. Perhaps two of
the most relevant techniques for this are audio and video mapping. In the
case of audio mapping one would record audio notes about various things to
map while traveling. Videomapping is even easier, as one would simply
videotape (e.g. with a helmet mount camera, or a web cam on the car
dashboard) a trip and one doesn't need to do anything at the time. However,
post processing this later on into map data can be rather tiresome and
boring, as one has to sift through possibly hours of traffic recordings to
filter out the interesting bits.
This is where the proposal for the video based speed limit detector comes
in. The idea is to automate as much as possible this sifting through the
video recording and make the life for the mapper as easy as possible to
extract relevant information.
In its simplest form, the project would be to go through a video recording
image by image and detect if there is an interesting street sign in the
video. If there is, it would pull that image out, correlate it with a GPS
track that was recorded simultaneously and present those images as photos in
JOSM to be used in the standard photo based mapping in JOSM. To make the
work-flow as simple as possible for mappers, the system should be well
integrated into an editor like JOSM. It looks like you might be able to base
your work off of the video mapping plugin for JOSM .
Once this minimal goal is achieved, there are nearly a limitless number of
extensions one can play and experiment with to improve the usefulness to
One thing would be to not only detect an interesting street sign like a
speed limit in a picture, but also to recognize which sign it is and then
allow the mapper to automatically apply the correct tagging for the sign.
Another idea would be to try and use the video stream, the speed of travel
and the perspective increase as the sign comes nearer to estimate a more
exact location of the street sign, to get the beginning of e.g. a speed
limit as accurate as possible. One can also spend a lot of time trying to
get the detection and recognition algorithms as robust as possible, so that
they can work in varying lighting conditions or also with cheap web cams.
Although the suggestion mentions speed limits, one could use other signs as
well. However, speed limit signs (at least the european ones) should be
comparatively easy, as they are designed to be very obvious with very high
contrast. Furthermore, they are fairly important to routing and so far have
been less common to tag in openstreetmap. As such missing speed limits is
one of the most common bug reports people report  with using OSM in a
With respect to preparing for the submitting of a proposal there are
probably two things you should do:
1) As Graham already mentioned, you should (if you are not already) get
familiar with the idea behind OpenStreetMap and mapping and how the current
editors work. The best way to do this is my doing some old school mapping of
your area you live in. i.e. go out on a dedicated mapping trip. This should
hopefully amongst other things get you a feel for what is currently
difficult to map and therefore what would be most useful from an automated
2) There is quite a bit of research literature on the problem of video based
speed limit detection for car navigation. It would probably be useful to
read a few of those papers, just to get a reference of what has already been
done outside of OpenStreetMap and what kind of ideas and algorithms they
If you have any questions or ideas, I'd be happy to discuss them.
Good luck for applying.
View this message in context: http://gis.19327.n5.nabble.com/GSOC2012-Video-Based-Speed-Limit-Detector-tp5583971p5584528.html
Sent from the General Discussion mailing list archive at Nabble.com.
More information about the talk