[soundfx]

My team added vision processing to our autonomous code between competitions to supplement our dual camera setup that was already in use at our first competition.

Cameras worked fairly well for our drivers, especially since we were able to add cross-hairs to the feed for easier alignment. However, it still took a bit of time to line up, although it did a great job of ensuring the shooting was on target. We were able to send over a USB camera feed and an image from an IP camera (for processing) with little trouble at events.

Our vision tracking worked twice throughout our second competition, though most of our troubles were rooted in our lack of carpeted area large enough to test on. We elected to process the image on our driver station laptop, and did encounter quite a bit of lag through the IP camera and network tables, but we were still able to make it accurate enough for two working shots in competition.
We did not have problems with the network (besides the lag) as long as we accessed the camera via its name, not IP address (e.g., axis-camera.local, not 10.32.38.11).
I'm sure our system could have worked from the defenses if our shooter had that capability.