This article was mentioned in Slashdot and elsewhere ... Wimbledon tennis is underway and new uses of image analysis to determine line calls, notably the 'Hawkeye' system is now in common use. Based on listening to the commentators in Tennis, the system's accuracy is seen as absolute. In particular the visual display created by the system is so compelling that there seems to be no room for dispute. Yet there are considerable errors in the system itself. How is this reconciled? It reminds me of a presentation I saw from a consultant which showed a very realistic visual display of atmospheric changes that would occur based on a complex simulation which used a number of positional and operational variables. The output was a realistic graphic which displayed a beautiful blue sky in one case, and a smoggy layer in another. So the output was 0-1 (in or out?) I asked the obvious question: What errors in the system could cause a flip from smoggy to blue sky? Had these been explored? What is the sensitivity of the model? I did not get a real answer, it was an executive briefing after all.
You could make the argument that the tennis game has become so fast that despite the systems errors it is still better than having a human judge the result. True, but I woud like us to make that choice in an informed a way as possible.
You can get a preprint of the technical paper by clicking on the link above.
No comments:
Post a Comment