Monday, January 21, 2008

Hawk-Eye ball-tracking technology

After a weekend of some remarkable tennis matches in the Australian Open, conversation turned to Hawk-Eye technology and similar techniques used in cricket.
Some of this discussion was a repeat of a conversation that Len and I had in the first week he was here (this year), following which he found this blog entry by someone claiming to have worked on the development of the Hawk-Eye system. He (or she?) makes some interesting points about the error(s) inherent in the system and the assumptions of the ball being spherical and hard when it hits the ground on or near the line.

I'm inclined to agree with Maverick's point (the fifth comment on that post), despite the original author's retort, that it doesn't really matter if Hawk-Eye is accepted as imperfect or not, if there is objective data that can show that it is statistically more accurate than human judgments. The reply that the Hawk-Eye error 'is biased and therefore that makes it worse' (paraphrased) seems to be suggesting that a higher proportion of 'called-in' balls (like the Federer-Nadal case cited) should actually be 'called-out', because they model the ball touch-point as a circle at the full diameter of the normal ball.

But, I don't buy this. If Hawk-Eye modelled the touch-point as a full-diameter circle, that's what you'd see. In fact, what you see in the Hawk-Eye replays is an ellipse (ie., they clearly do model ball compression, taking into account the trajectory in their calculations).
You can check out the Wikipedia entry, where it suggests that the average error of Hawk-Eye is about 3.6mm, or 5% of the ball diameter. The Wikipedia entry author makes the point that this is approximately equivalent to the fluff on the ball. Or even better, the Hawk-Eye website itself, where you can find this document explaining the controversial line call - the sequence of 10 pics at the end is particularly interesting in this context.
So Hawkeye isn't perfect. But, on average, overall, it's better than people for the particularly close line calls.

No comments: