[OSM-dev] [GSoC] Video based speed limit detector

Kai Krueger kakrueger at gmail.com
Tue Mar 22 07:33:44 GMT 2011


great to see that someone is interested in the project idea and I'll be
interested to see the proposal draft.

There are a number of standard object detection algorithms that should be
able to detect speed limit signs on a frame by frame basis in real time.
Once a sign is located in the image, one then has to interpret it. One can
either try and do a full OCR style approach, or given the limited number of
speed limit signs, one can probably simply throw a multi class classifier at
it with a class per speed limit. These algorithms are reasonably robust, so
they should be able to cope with the difficult conditions of changing
lighting conditions and size of the signs in the images. (After all traffic
signs are designed to be well detectable and readable)

For both aspects there are ready made libraries like the OpenCV library that
implement most of the algorithmic side of things. So I am reasonably
confident that the minimum goal of analyzing a video stream and identifying
the speed limits, spitting out the gps coordinates of the video frames for
those signs is achievable in the time scope of GSoC.

My guess would be that quite a bit of the initial project would be to
experiment with various algorithms to see which work best and combine them
in ways to make it as robust as possible. 

Once that basic goal is achieved, there are nearly limitless (at least with
respect to a three month project) options to extend the project in various
directions, either be it from a usability perspective, or from an
algorithmic computer vision perspective, giving it the flexibility to be a
challenging and interesting, yet achievable project.

Furthermore, I think one can achieve something in the time that is genuinely
useful to the project and helps improve an important aspect of tagging for
routing that has so far been comparatively neglected.

I possibly still have some code lying around to interface the C library
OpenCV with java through JNI from my undergraduate thesis (which was to
build an eye tracker from a standard webcam), in case that would be of any

So I am excited to see someone willing to pick up the project.  

I'll possibly be able to mentor such a project, although I know little about
the code of any of the editors, so I'd be less able to help on that side of


View this message in context: http://gis.638310.n2.nabble.com/GSoC-Video-based-speed-limit-detector-tp6192971p6195251.html
Sent from the Developer Discussion mailing list archive at Nabble.com.

More information about the dev mailing list