[Optimoz] gesture recognition with a hidden markov model

[Jens Bannmann]

> Specifically, the file must contain:
> 
> 1) All valid output strings (sequences of characters from {U, L, R, D,
> 1, 3, 7, 9, *})
> 
> 2) Probability estimates for each of these strings.  For now, these
> don't need to be perfect, but we should try to make them reasonable. 
> In the future, we can probably run a modified MozGest that logs
> gestures, and then we can combine all the data to get empirical
> estimates.

Incidentally, MozGest already logs usage counts for each gesture since 
version 1.0. Currently, they are solely used for ordering the gesture 
legend sidebar, but the long-term idea has always been to let users 
voluntarily submit their usage data (in anonymous form) to a central 
server. There, we could create all kinds of statistics, and e.g. decide 
which gestures should be removed from the default set.

But even lacking such a feedback system, we could use the data today.

If data from only a handful of users would help you, we could have 
mailinglist subscribers extract the data and send it to you. The problem 
would be that it would not include gestures that particular user doesn't 
use, or list 0 as the probability for them. Additionally, there's no 
such thing as "all valid gesture strings", only "all valid gesture 
strings the user has configured". If both limitations are okay for you, 
let me know and I'll post the explanation how users can extract their 
data (an XSL transformation stylesheet will come handy).

Oh, and here's a random part of my usage counts, manually converted into 
your format. Let me know whether this would work.
*RUL 20
DL 4
DR 792
UDU 1
URD 103
L1D3R 2
7 8

Regards,
Jens