Dave the Funkatron said the following on 05/06/2008 16:12:
> On Jun 5, 5:53 am, Richard Brooks <richardbro...@[EMAIL PROTECTED]
> armstrongs.com> wrote:
>> Dave the Funkatron said the following on 05/06/2008 03:00:
>>
>>
>>
>>
>>
>>> 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 ****ge 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
>> Maybe some kindly student soul could take the data, im****t 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.- 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,
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.
|
