I used to work on the system, so I can answer (some) questions about it.
There is a nice article in wired about it(include a picture of one of the cameras and the view from another) http://www.wired.com/playbook/2011/04/nba-data-revolution/al...
The calibration is done by matching(automatically or manually) known world points to pixels from the cameras. (This is a very common practice in image-processing systems)
I was looking at espn's feed of the game the other night and they seem to have a lot of details about the game as the game is being played. things like "player missed 13 foot jump shot" and a graphic of a ball bouncing off of the rim and where it was shot from. Id imagine that they have a system like court stenographers so that they can quickly record info and send it to their servers.