Hello fellow gamedevs,
i have a quick question regarding gesture recognition and hope for some of you to be better informed than i am. (Which is most likely the case.)
What i did first:
Me beeing new to the field at first took a very naive approach, sampling my sensor x times per second and then did a likelihood analysis (via a Levensthein adaption) - which was painfully slow and not really robust (for obvious reasons).
What i do now:
My new approach differs quite a bit of that.
Now i have a list of sensor data, a time value how fast the state has to be reached to be satisfied and a fuzzyness factor. Every samplepoint in my list then basicly acts as a state and i store which state was last satisfied.
A state is satisfied, as soon as its predecessor is satisfied (state -1 is always) and we reached its ancestor in the given time without leaving the capsule spanned by the two samplepoints and the fuzzynessfactor(as radius).
The question(s) beeing...
Does this approach have a name so i can do some readup on that? I am not very well acquainted with the terminology as many of you might have guessed by now, so i couldnt find anything so far.
Also: Does this even sound like a feasible approach? It works for very simple gestures and i am currently in the process of testing more advanced ones, but it also clearly has its shortcomings...
I hope someone is able and willing to help, if you need more information i will provide this as soon as possible. Thanks in advance and warm regards,
PS: I almost forgot to tell! My sensor data is continous and clamped (since its angular data). Positional data is also planned but should work similar.