Thanks to everyone for all the responses. I’ll clarify a bit on the questions asked:
Did you use the default values for the green light of modified values?
We actually tested both the default values and a set of modified values. When at home, we found small fluctuations in the green light’s color and made modifications accordingly. When these values didn’t seem to work at the regional, we went back to default values and found less interference from the arena’s lights…however, not enough to call our camera “effective”.
A simple question perhaps but did you make sure it was focused?
Actually, this was a big problem in our camera missing the light during our build season. This was one of the first things we adjusted at the regional events. The camera was able to track the green light on the practice field (where the arena’s lights were absent), but caught lots of interference on the “game” field.
One question: Did you find it working great tethered and then horribly on the game radio?
We had that problem at Boston. Our radio stopped working correctly right after UTC. The robot worked great on the practice field while tethered, but then ran unpredictably on the field. Once we realized it was the radio and put a new one on, we were back in business.
You might want to try it. Goodluck!
This was something we did NOT try at the regional. We assumed that the radio was functioning correctly since it worked well at home. But then again, we thought the same thing about the entire camera module. Thanks for the tip…it’ll be good insight to take into next year 
The lighting in the arena could confuse the camera, this becomes very clear when you see the camera lock on to one of the lights. We saw problems in our shop environment with fluorescent lights where the aim of the camera would be off and the confidence value would fluctuate. When we turned the lights off the problem would go away, so we knew it was the lights.
We addressed this issue as follows:
-
Track the confidence value in your code. If bad confidence comes up use the most recent data report from the camera with a good confidence. Our cutoff in this regard was a confidence of 100, bad values were perhaps 30 or less, and typical good values were 160 or greater. When the camera was being confused by the lights, this confusion would fluctuate frame to frame and the ones that were good were marked by good confidence.
-
Don’t use automatic search. Use operator control of the turrent to get it pointed in the right direction and have the operator flip a switch to release control to the computer when the green light is seen with a confidence above the cutoff, 100. Your turrent looks a lot smarter to the judges this way. All you need do is indicate good confidence on the OI with an LED.
-
In spite of this, we still had problems at the San Jose regional. This problem was solved by putting a blinder on the camera that kept the arena lighting out of it.
Our shooter did not tilt and the camera was hard mounted to it. The fact that the camera did not tilt probably made it easier for us to construct a blinder that worked. It was cut from cardboard right at the regional and stuck on with velcro, after blackening it with a sharpie pen. It worked really well.
Eugene
This sounds similar to our problems at San Jose (and Davis for that matter). At a point during our build season, our programming team noticed interference in the camera coming from the fluorescent lights in our auditorium. Hence, we actually use a similar method to light-tracking…using the “confidence” data packet to determine what type of light was seen (confidence value of 80…which pretty much eliminated our fluorescent light problem).
As for turret vs. camera, our camera functions completely independently of the turret. It works on its own pan/tilt axis (via the provided camera-servo mount) and our operator decides when the turret should auto-adjust to the camera’s values. We did experiment with polarized sunglass-lenses over the camera, but not to a large extent.
The blinder is an interesting idea; we never thought of doing something like that. Our camera does require a lot of freedom of motion in its current design however, so the “blinder” would not essentially “blind” us much. Maybe we tried to do too much with the camera?
One thing I wonder is: Could the “color values” of the green light be slightly different ON the field given that the field is brightly lit with “arena-style lighting” while all background is completely dark (or backlit by other lights)?
Again, thanks to everyone for the replies…they really help us think through this problem.