I am interested in how this study progresses, and must admit I'm more into gestural/temporal analysis than the hardware side of things (though it is nice to keep abreast of the latest developments…). Especially ideas that could also be applied to music/movement coordinations.
Definitely agreed that the simple approach of onset/intervals is the way to go with this one. I just wanted to note that, in such unconstrained movement situations, the continuous signal is still an important friend when one
cares about defining segments or onsets. In Max/MSP parlance, bonk~ is right for some situations and not for others. Conversely, continuous cross-correlation can be an excellent tool for finding onsets and the lag between co-ordinated onset actions, or a misleading one depending on the ensemble of signals. Really it should be tried, though along with the inverse, discrete correlation of pre-extracted onset data. Weighting, warping and normalization of the corr. function can be applied as the situation receives more constrains due to the movement context. (e.g. compare Yin algorithm to standard autocorr. in the case of pitch detection).
All this to say, allowing a pre-defined package to uncritically handle the layer between continuous input and output onset data could be another type of over-determination of the problem : )
best,
Doug