Motion following with HMM 
Friday, November 26, 2004 18:49
The goal of my master thesis was to generalize the idea behind ircam's score follower to the field of dance. The system is based on a video analysis of what is happening on stage done with eyesweb, which is compared to an Hidden Markov Model based on other realizations of the same choregraphy.



Documents


you can download my master thesis
and also browse the bibliography online.

I've made my best to cite references correctly, however I wasn't able to collect all the necessary information for some papers. If something is cited incorrectly please tell me.


a sample application


using the motion following, we can synchronize a live dance sequence to recorded one.

Error text.
video A1 performed by a dancer

Error text.
video A1b, the same choregraphy performed by another dancer, with a slightly different timing


Error text.
in the upper part: choregraphy A1
below: frames from A1b synchronised to A1.


Obviously there are jumps inside the video because the 2 dancers don't perform exactly the way.

More details




4 executions of the same choregraphy performed by 2 different dancers. The trajectory is drawn in the space of the first 2 principal components.
Colors and point styles represent different steps of the choregraphy.