Overlayed crests, scores and text on pitches

Predominently on sports programmes the broadcasters electronically superimpose crests, scores and text on pitches and sports grounds.

I know about Chromakey and CSO but on these superimpostions players, horses etc actually seem to 'walk' all over the superimposed image too. It appears to be a CSO on a CSO (If that makes any sense)

Does anyone here have access to a webpage describing how this technique is actually done or can describe it in this thread.

Thanks,

Chris.

Viz|Arena is one of the hardware/software solutions that enable broadcasters to do this.

You can see details about it at this website.
http://www.vizrt.com/db/106/7/23/document11.ehtml

It's just a normal CSO (albeit with a very clever virtual gfx infill). In any CSO setup, something moving in front of the blue or green background will block the key colour in that area and suppress the key - think of a weather presenter in front of their CSO-keyed map. That is, of course, unless the thing contains some of the key colour in itself, in which case that bit of it will appear transparent - e.g. somebody wearing a blue or green tie would appear to have a hole the size and shape of the tie in them, with the background showing through it.

This was seen on Look North Hull that time with Paul Hudson's shirt coming see-through.

The best picture is at the bottom of the page from the link:
Paul Hudson's Shirt Mishap

The graphics are superimposed using regular (though high quality) CSO/Chroma key techniques - it is helpful that the grass on the pitch is usually a uniform green meaning you can usually get a decent clip. On swimming it is a bit more difficult - though the water is often blue enough.

The camera is tracked using a number of systems - some (Epsis for example) use a fixed camera which is accurately described in position terms (using laser theodolites) relative to a simple 3D model of the stadium and pitch, which has a sensor pan/tilt/zoom panning head that allows these to be fed back to the graphics generation device. Other, newer devices don't need the accurate position or tracking head, and use scene analysis techniques (often using the white lines on the pitch) to detect where the camera is pointing.

Generating the graphics themselves is quite easy - it is the camera tracking that is the really clever bit, and the keying is quite specialised, though not amazingly so.