advice needed on shape recognition- cv.jit.learn- cv.jit.undergrad ?

hi ma()xists,

i am working on a new theater project involving shape recognition. Actors will do certain gestures that need to be recognized by Jitter. The gestures are fairly different from each other. when a gesture is recognized a movie is triggered.
right now i have pictures of the gestures that will be compared to an incoming videostream of the same actor who has been photographed redoing these same gestures ( lighting, shooting angle are the same in the pictures and the video )
i started working with the cv.jit objects.
i'd like to to know what would be in your opinion the most efficient way to deal with shape recognition.
i was planning to work with [cv.jit.learn] or [cv.jit.undergrad]. any advice on training these objects would be welcome.
has anybody done such things successfully ?

thanks in advance.