Focusing Solved

After some more research and advice, I have managed to get the system working much more smoothly and consistently.

The first issue was the focusing: after some good advice from Adam (comment here) I flipped the frosted glass table top over so that the frosting was closer to the tag patterns. The difference was instantly noticeable: no more blurry tags!

The other mistake I had been making was assuming that the green dots marking the tag centres in debug view were a good thing, so I was adjusting the threshold control until all these dots turned green. It turns out that the brighter the green, the less certain the software is of the tag's position - once I realised this and adjusted the threshold the other way (until all the centres were dark), the tracking accuracy improved dramatically.

I am also running the table without its internal lighting at the moment; just boosting the gain on the webcam slightly and relying on the room lighting coming through the table top. This seems to work well, and the tracking works much better over the whole surface, which I think is due to the lighting being much more even. I can now experiment with some different ways of diffusing the internal lighting, or perhaps use a number of LEDs, to make sure the surface is receiving roughly the same amount of light all over.

I will try and get some video up as soon I can get hold of a video camera!