Shop OBEX P1 Docs P2 Docs Learn Events
Project involving detecting road signs and traffic lights — Parallax Forums

Project involving detecting road signs and traffic lights

Bulslayer85Bulslayer85 Posts: 34
edited 2012-04-18 13:01 in General Discussion
My robot project is getting a "smart" motorcycle to recognize different road signs and traffic light colors....let alone the light its self. The idea is to alert operators to the presence of important signs and changing of lights incase the person is un-aware of it already. would i use a video feed back and how hard would it be to program the robot to compare different images?

Comments

  • Phil Pilgrim (PhiPi)Phil Pilgrim (PhiPi) Posts: 23,514
    edited 2012-04-16 10:41
    Bulslayer85 wrote:
    ...how hard would it be to program the robot to compare different images?
    It would be Ph.D. dissertation-worthy.

    -Phil
  • kwinnkwinn Posts: 8,697
    edited 2012-04-16 21:30
    Phil Pilgrim (PhiPi) wrote: »
    It would be Ph.D. dissertation-worthy.

    -Phil
    And then some!
  • GadgetmanGadgetman Posts: 2,436
    edited 2012-04-17 02:11
    Actually, the car manufacturers are demoing prototype systems already...
    But I think they mostly focus on speed limits and right of way. Even then, they have big problems just recognising what is and isn't a sign, or even if it's of any concert to the driver.
    (Sometimes there are parallell roads so closely placed that there's only a ditch between them. Here in Norway, a 'speed change' or change of right of way status is done with signs on BOTH sides of the road, not just on the righhand side. The sign on the lefthand side may then be close enough to the righthand side of a parallell road to be mistaken by some as to belonging to the other road. Being able to see another to the far right though, will reveal the truth.)

    What about old and worn signs?
    (Some signs can be faded almost completely, but most humans still manage to 'rebuild' it in their mind)

    Signs that are damaged?
    http://en.scanpix.no/spWebApp/preview.action?search.offset=5&search.rid=1112624&search.rbase=SP_NOR_02&search.searchId=1454708940&search.previewNumResults=200&search.tabId=editorial&search.advanced.dateCreated=1997-07-25
    (Bulletholes in 'moose signs'... )

    You don't have to think about signs covered in snow, though...
  • prof_brainoprof_braino Posts: 4,313
    edited 2012-04-17 06:07
    Bulslayer85 wrote: »
    My robot project is getting a "smart" motorcycle .... would i use a video feed back and how hard would it be to program the robot to compare different images?

    I would love to work on something like this as part of a team. Its obviously impossible, so we should have something working by the end of summer :)

    My initial input, as always, is 1) the degree of success of any project is directly related to the degree of planning; and 2) everything starts and ends with requirements.

    So, can we organize in detail exactly what is expected/required? The intent is describe the system(s) in enough detail (in advance) so that a piece of code can be written to perform that function. OF COURSE there are many unknowns, and many of the initial requirements will require some investigation and research, to further clarify what is being asked, and what results can be expected. I can help with this, I do it all the time with great success.

    But just because it's PhD dissertation worthy does not mean its a waste of time to pursue it if you are shooting for a PhD. The a very valid reason to do it would for the selfish reason of "because I want one, and nobody would sell me one".
  • skylightskylight Posts: 1,915
    edited 2012-04-17 06:13
    If they can do it with face recognition then why not?
  • LeonLeon Posts: 7,620
    edited 2012-04-17 06:41
    They are both difficult problems.
  • mindrobotsmindrobots Posts: 6,506
    edited 2012-04-17 07:43
    Do some Google searching for Sebastian Thrun and the Google self driving car. There is a lot of information on the technology supporting this effort and where the "state of the art" is at the moment. It might give you idea of the scope of the problems and the road to the solution.
  • skylightskylight Posts: 1,915
    edited 2012-04-17 07:53
    I would have thought that face recognition was far harder as the face can be pulled into various shapes and also angles whereas signs will always be facing the camera and be standard patterns so searching a database should be far easier? Perhaps to make it easier all signs could be fitted with barcodes.of some type ie a special pattern for the camera.
  • TubularTubular Posts: 4,709
    edited 2012-04-17 23:14
    I was watching a demo of some high end PC based vision software recently. It showed the ability of the software to detect road signs, as well as reliably detect obscured and mangled logos (ie should work fine for deformed signs). I'm impressed with how far the PC vision software has come.

    I would think OpenCV would be a good place to start. I believe it can run on the BeagleBoard (or similar), and perhaps Android, which would ultimately let you create an 'App' for your dash mounted phone. Combine that with GPS sensing of speed and you might have something useful.

    I assume OpenCV cannot yet run on a prop or prop2 (no matter how slowly). I would love to be wrong about that.
  • mindrobotsmindrobots Posts: 6,506
    edited 2012-04-18 07:18
    Hanno Sanders' Viewport IDE has OpenCV support built into it. There is some doccumentation on his site (hannoware.com) and there are some videos with him domonstrating the propeller/OpenCV connection. This is a start as to where the Propeller and image processing with OpenCV is.

    Something to check out if interested.
  • LoopyBytelooseLoopyByteloose Posts: 12,537
    edited 2012-04-18 09:22
    Start with color, then shape, and finally with the text or iconic image. Green signs are informational and can be easily ignored. I believe that only Stop signs are red. At least in the USA, it is rather well thought out. But here in Taiwan, you would be likely to confuse all the advertisements with the traffic signs. And there is always the problem of someone having a billboard with a Stop sign included in the image.

    It would all be much easier if ALL traffic signs included a scanable bar code. That would make a lot more sense. The laser bar code image could be put on the same pole, but at the height of hub caps so lasers don't strike the human eye while scanning.
  • xanaduxanadu Posts: 3,347
    edited 2012-04-18 13:01
    This one is pretty cool... http://www.youtube.com/watch?v=hU7yHQkg-7U

    There is CCTV software that can detect objects. When it detects the object(s) you can predefine an alarm. You'll need a pretty good CPU and memory so mini-itx at the least. Then you'll really need a good camera, and I'm going to guess it would only work at really slow driving speeds, like 1-2mph. It might be a step in the right direction. That's a huge project though.

    If you want to play around with image detection http://ge.ecomagination.com/smartgrid/#/augmented_reality there is a good example of the performance you can get on a very sound setup. You will see that even inside while you're not moving it takes quite a bit to pickup on a high contrast marker.
Sign In or Register to comment.