They improved on the Microsoft implementation with their algo called KinFu Large as they are able to scan multiple volumes in on pass allowing to scan larger scenes in one go.
The point cloud library (PCL) is available as prebuild binaries for Linux, Windows and OSX as well as in source code from their svn repository. The code relies heavily on the NVidia CUDA development libraries for GPU optimizations and will require a compatible GPU for best results. Information on how to setup your own build environment and the required dependencies is available from their site.
The preliminary source code is currently available in their SVN repository’s. The code relies heavily on the NVidia CUDA development libraries for GPU optimizations and will require a compatible GPU for best results.
Besides the Kinect the library supports several other sensors. Moving forward, the developers want to continue to refine and improve the system, and are hoping to improve upon the original algorithm in order to model larger scale environments in the near future. The code is still beta, a stable release is planed to coincide with the upcoming PCL 2.0 release.
I’m definitely looking forward for to what the Kinect community is going to do with that.