kinect + avatars

[Deltcho]

I recommend using the official microsoft SDK for the Kinect (it's free). The Kinect tracks very specific points on the body and you need to interpolate the position of the body from those points. The "Skeletal Tracker" demo that comes with the SDK should be all you need to get going. Just integrate a vrpn server into the code (and remove the visualization stuff to improve performance) and you're ready to go.

Or, you can calculate the position of each limb in euler coordinates within vizard using FAAST, but python is significantly slower than c#/c++ so I'd recommend doing it within the SDK framework and just sending the limb position over VRPN to Vizard.

When you've got the coordinates for the appropriate avatar limbs, you just link them within Vizard. It's a shame that Vizard can't work with the Kinect out of the box using official drivers.

[tobin]

Thanks for the comments about the MS Research SDK. The lead developer of the FAAST project, Evan Suma, offers the summary below of the pros and cons of the MS Research SDK and the OpenNI SDK (note that FAAST currently uses OpenNI but will soon add support for MS Research SDK thus giving users the ability to choose):

"OpenNI advantages:
- More accurate and robust for arm/leg pose.
- Better depth map resolution (always 640x480). When the skeleton tracker is activated with Microsoft, the depth map drops to 320x240.
- More permissive license that allows corporate use. The Microsoft SDK can only be used for research and noncommercial purposes.
- More stable. The Microsoft SDK is a little more buggy and unstable, and crashes for me every now and then.

Microsoft advantages:
- No calibration required. However, the word from PrimeSense is that they are very close to solving the calibration issue, and hopefully soon OpenNI will no longer require the calibration pose at all.
- More skeleton joints. With Microsoft you get an articulated neck joint, which is rigid with OpenNI. You also get the wrist and ankle joints, but in my experience these don't work very well.
- More sophisticated built-in filtering (to reduce noise and jitter).
"