He estimates a throughput of 30 seconds/day which is probably not much less as what it takes to do manual cleanup for normal motion capture.
Via https://vimeo.com/groups/ice/videosTest of My ICE based manual mocap system using cheap consumer cameras.
Taken of a Llama in a field from a relatively narrow 3 camera view. ( the quality goes up, the wider it is (and the more cameras))
No real animation skills required. ( just decent spacial awareness.)
It is manual, but I estimate that 30 seconds of this could be done in one day. That's a lot faster (and more realistic) than I could animate it.
I have plans to make a semi-automated version in the future.
Would anyone be interested in Kick-starting me making this into a tool/plugin for Maya?