Cleaning up motion capture data

[Dave the Funkatron]

Hey all,

I have some general questions about how to turn raw optically captured
marker data into an actual skeletal animation. It seems like several
problems can pop up along the way. This is all probably elementary
stuff, but so I think I just need a nudge toward some tools,
techniques, research papers, etc.

First, the skeletal segments are generally treated as rigid, but the
human body (at least the flesh part) is rather deformable. So, markers
tend to slide around on the body, if ever so slightly. So, if I try to
fit a rigid skeleton to that data, I won't be able to find an exact
solution. Well, I can get a best fit, but the fit between adjacent
poses in the time-line might be quite different and so I would get
discontinuities in the animation.

Another issue is that I might actually lose a marker or two along the
way, and have to fill in the data along the way. It seems common
enough to interpolate in these cases, but that also has an effect on
the best fit, and might cause more discontinuity.

I'm sure that there are other problems that I have not even thought
of.

It seems like people would have run into these problems before, though
I can't seem to find any research papers on the subject. I have been
told that tools like Motion Builder can do some clean-up, but is not a
bullet-proof solution. So, can anyone suggest alternatives or places
to start looking?

Thanks.

Dave

[Richard Brooks]

Dave the Funkatron said the following on 05/06/2008 03:00:

Maybe some kindly student soul could take the data, import it into a
'home brewed' program which can do some mathematics on the data? It
shouldn't be hard to do as audio editors do the same sort of thing
with low-pass filtering and click removal. The data can then be
written back out to a file in the same format as the original and tested.

[Dave the Funkatron]

On Jun 5, 5:53 am, Richard Brooks <richardbro...@vickers-

> written back out to a file in the same format as the original and tested.- Hide quoted text -
>
> - Show quoted text -

Well, low-pass filtering can fix pops that are only a frame or two in
duration, but they also remove other details, and they don't fix long
durations of this sort of artifact,

[Richard Brooks]

Dave the Funkatron said the following on 05/06/2008 16:12:

But all kinds of equations can be written and over any number of
frames. I'm having to do that for my own 4-channel IFL writer with
convolution.