I made a similar product (albeit a lot less refined) for a hackathon, using the Pebble. It counted number and type of tennis shots, displayed useful HUD data, and then let you see pretty graphs in a desktop webapp. Some problems I ran into:
If you have acceleration, to get position, you have to integrate twice, which means two unknown constants, really throwing off your data accuracy. To fix this, the user would have to go back to a defined position (perhaps the tennis 'ready' position) frequently, but that isn't guaranteed or user friendly. They could absolutely work around this with some refined heuristics, but I wasn't able to get one working in the few hours I had.
The algorithms are the hard part. I was using naive machine learning techniques to identify swings and such, and they were not nearly adequate; I had many false positives. If they can get all of the stats they claim, I'll be very impressed.
If you have acceleration, to get position, you have to integrate twice, which means two unknown constants, really throwing off your data accuracy. To fix this, the user would have to go back to a defined position (perhaps the tennis 'ready' position) frequently, but that isn't guaranteed or user friendly. They could absolutely work around this with some refined heuristics, but I wasn't able to get one working in the few hours I had.
The algorithms are the hard part. I was using naive machine learning techniques to identify swings and such, and they were not nearly adequate; I had many false positives. If they can get all of the stats they claim, I'll be very impressed.