I’m working on an independent study this semester of school with the goal of setting up a mobile motion capture studio and pipeline on little to no budget. Sounds fun eh?
My ideas so far involve capturing skeletal animation using a kinect and OpenNI, getting it into MOBU in a workable file format, Retargeting to a character, exporting that retargeted data into Maya, cleaning up the animation and animating the hands, feet and head, and finally exporting as an FBX to be imported into UDK for cinematic animation.
I’m trying to accomplish as much of this as possible using python, as I’m (for now) a newb at writing code. I’m currently in the research and pre-production stage of the project, and compiling a list of libraries and options.
I’ll be updating this thread weekly as the project progresses, to both document my work, and hopefully get feedback and critique on my process.
I’ll start off with an open request for critique and advice. Are there any Libraries or sites or projects that you guys would recommend?
My list of topics to investigate so far:
OpenNI
PyOpenNI
OpenCV
PyGame
MotionBuilder’s python compatibility
General File handling and formatting
I’m extremely open to any advice or critique, so fire away!
A month ago, our studio used kinect to make pre-visual for cinema.
check this:
Brekel kinect http://www.brekel.com/?page_id=155
Brekel’s kinect plug-in for Motion Builder works in realtime but too much jitter…
Ipi soft http://ipisoft.com/
The software from ipisoft works great. It does not work in realtime but the quality was pretty good.
We could use multiple kinect hardware(we used 1 Xbox Kinect and 1 Asus xtion pro live) to increase accuracy.
There was a talk at PyCon this year that covered some topics that might be relevant, namely python, a kinect, and OSCeleton as an assistive technology stack:
For Python and OpenCV, you have a few choices, i’ve been bouncing back and forth between the native OpenCV wrapper and SimpleCV, definitely worth a look as far as getting up and running with CV. There are a few perf issues with SimpleCV, but nothing showstopping.
If you don’t mind switching domains a bit, I’d actually recommend simple-openni for processing to learn the basics of OpenNI. If you’re reasonably comfortable with Python, processing’s a breeze. Making the transition to PyOpenNI from there should be pretty straightforward, though, speaking from experience, it DOES become frighteningly easy to get caught up making cool Kinect-based visualizations as opposed to doing something useful :laugh:
Lastly, to cop a thought from the ever awesome Zach Lieberman, welcome to the front lines.
