|
[DPRG] Re: Games
Subject: [DPRG] Re: Games
From: David M Wilson
davidmw at tx.rr.com
Date: Wed Jun 20 09:43:22 CDT 2007
>>This gets on to how at least some real world vision systems actually work.
One way around the problem is to have your own light. Some systems that
appear purely passive actually have IR illuminators to try and mitigate
the effect of ambient light.
Chris,
Many thanks for the comments. A few weeks ago I discovered that the vision
system could return images of a classic 'baseplate' compass and that heading
information could be computed at about a 1 degree resolution. The red
needle is behind shiny plastic and would often fall into shadow or behind
reflected glare. Navigating with this compass is a bit like line following
but the task is to maintain a specific angular displacement of the target
instead of seeking to eliminate it.
I did find 1/2 the solution in providing a reliable light source - using a
white LED worked perfectly.
The other 1/2 of the solution was providing a reliable shade source that
could be carried with the robot. After trying antiglare film on the compass
face it was pretty clear that external light needed to be blocked. A simple
cone made out of foam sheet worked really well.
For this conversation the reflected glare on the compass face is exactly the
same as the glare off the floor. It is easy to underestimate how much glare
needs to be handled at this altitude and camera angle. Flat black mini-T
courses and carpet can bounce light with the intensity of a good mirror.
Perhaps the shade needs to be extended to the floor.
It should also be mentioned that the auto white balance feature is important
in environments with mixed types of lighting. When the bot views a red ball
from a different angle the camera adjusts to keep the reds red. But at the
same time the largest 'off white' object or patch of reflected glare can
become the new white reference for the camera. While these adjustments are
being made the resulting white blob target is changing rapidly and helping
the robot simulate a seizure.
Removing the IR lens filter or turning off camera calibration features would
make the camera pretty useless when facing the windows in the space. Line
detectors may be the realistic solution.
David Wilson
More information about the DPRG mailing list
|
