Team 341 had an awesome robot in 2012, and they graciously released their image processing code for others to learn from (see this
white paper). Of course, as its a SmartDashboard extension, it's written in Java... and I happen to prefer python myself, so I ported the generic image processing portion of their code to Python 2.7+OpenCV.
While not fully tested, this properly recognizes all the images that they provided in the zip file in the white paper URL. I leave it as an exercise for the reader to import pynetworktables and ship the computed data off to a robot.
I didn't see a license file with their code, so I presume it is distributed under the same license that the SmartDashboard/WPILib is released under, as is this code. I am not associated with Team 341 in any way.