[Optimoz] gesture recognition with a hidden markov model

[Aaron Lenfestey]

> (but I'm afraid the probability-of-given-gesture approach may be a total
> failure in this case. There are gestures for which my usage goes into

Just a followup.  Maybe it was a total failure in the end, in the
sense that with the current set of parameters the probability of a
gesture needs to be ~1/million before the recognizer starts making
obvious mistakes.  You guys can test this by screwing with the
dictionary.  This seems to imply that the probabilities we feed it are
almost irrelevant, and in fact feeding it a uniform distribution over
the gestures seems to work just fine.

Also, I did some benchmarks and recognition of a ~100 point data
sequence takes <10ms on my 3.8ghz box w/ gcc-mingw.  I've included
some functions in the API that allow us to do this computation
incrementally as we get data from the user.  This turns out to work
pretty much optimally, in the sense that there is an O(1) amount of
work to be done after the last data point is received.  Not that I
possess such a machine, but I expect performance to be acceptable even
if its CPU were 10 times slower.

-Aaron