|
[DPRG] Re: stella's test drive outdoors
Subject: [DPRG] Re: stella's test drive outdoors
From: Chris Jang
cjang at ix.netcom.com
Date: Mon May 7 14:17:30 CDT 2007
>It is hard to tell from the video what stella is reacting to.
>Are their boundaries or walls that it is seeing anc backing
>away from?
Hi David, the feature threshold was set too low. Ruts in the
pavement sporadically appeared, causing Stella to stop and
back up. The painted lines were clearly visible - except that
sunlight glare off the surface formed a featureless corridor
to drive through. That's how she was able to cross all of of
those painted lines. But she kept stopping as the lines would
again sporadically appear. Everything Stella sees is recorded
to a SD flash card so is relatively easy to analyze later.
I think I'll try something more obvious like driving down a
path or avoiding obstacles in a few days. I feel there is a
need for an additional filter stage in the vision pipeline to
reject more of the false positives.
>For the IDSC event, could you home in on a brightly colored
>target (say, a cone-shaped target that's internationl orange)?
Without testing, I don't know. Stella is presently color-blind.
She sees only a luminance image. To just get the webcam working
outdoors, sunglass lenses had to be taped over the camera
objective. So I'm not confident of the chrominance image due to
camera exposure problems.
My recollection is that orange traffic cones are highly visible
in chrominance but may be subdued in luminance. The only way to
know whether Stella can reliably see them, in chroma or luma,
is to test under realistic conditions and find out. Chris
More information about the DPRG mailing list
|
