Re: done any image recognition using the Logitech Quickcam SDK ?



>> As for your question as to code for matching
>> images that is really a question for an algorithm.
>> And that depends on what kind of images and
>> what kind of "match" you are talking about?


pogo wrote:

> The kind of "match "I am looking for is this:
> 1) I take a snapshot of some image, let's say
> a doorway, and store it in a database/list of
> some sort as image # n.
> 2) Then I want to be able to have a robot
> compare images (frames) it sees as it navigates
> a room against those in the database.
> 3) When it sees an image & determines with a
> certain degree of confidence it is a "doorway",
> it will react accordingly.
> So basically I am looking for a function to
> call and say "how likely does your current
> image match image[n] ?". If it is easier for
> whatever reason - to match images of barcodes,
> or signs, or some kind of standardized icons -
> that would help me achieve the same purposes
> since I could just place those signs/icons on
> the doorframe, etc.
>
> This is more or less what the Evolution Robotics
> ER1 robot does and I am looking for a way to
> achieve the same functionality.
>
> Perhaps if nothing else those reading this post
> can help me ask the right questions based on
> what I have written above ?
>
> Thanks for any and all help !

If you are happy to structure your environment
with signs and icons that would be the easiest.

The method I used was to do a local threshold
on the image to reduce it to a collection of
blobs which could then be compared with a list
of blob shapes in memory. High contrast signs
will stand out against the rest of the image.

I used a white card with a black shape but you
could also use a color shape to take advantage
of a color camera. Just make sure the background
will contrast well.

Look up blob analysis sites on the web.

The webcams are noisy so you might precede the
local threshold with some smoothing operation
such as a median filter to get rid of the salt
and pepper noise.

To locate a door you might just put a vertical
strip which would contrast with the background
on say the handle side. Perhaps have the door
handle at the centre of the strip so the robot
know where to locate it. The size of the strip
in your image would indicate distance.

You can try all this just by taking some images
with a digital camera and using VC++ to load the
bitmap which you can than process.

As I indicated I played around with this stuff
with DOS and C but was never able to get the
info I needed to duplicate it with VC++ to
take advantage of the latest color cams so I
am unable to give you VC++ examples.

--
JC

.

Relevant Pages

  • Re: minimum print time test
    ... test use a clean negative to cover half opening of the ... make a serial of strip on the paper. ... exposure and contrast that result in good looking mid gray ...
    (rec.photo.darkroom)
  • Re: poly vs. nitro clear coat - age
    ... (finish, test, strip, finish, test), played by a freakin' robot (to make sure all notes are hit the same way). ...
    (alt.guitar)
  • Re: Why do so many SF fans think "Optimistic" = "Utopia"?
    ... want your daughter to hang out with" dude from the docks when the Oyl ... Robotman dropped the robot. ... The strip now has ANOTHER robot though, who came along with the old time- ...
    (rec.arts.sf.written)
  • Removing HTML Tags from a varchar2 field
    ... I'm trying to take some HTML fields and strip out the HTML ... formatting. ... The Policy_Filter seems to want to use a Blob, ...
    (comp.databases.oracle.server)