image processing toolkit and training data suggestions

Hey all,

I am interested in writing my own skeleton tracker for kinnect data, but I don't want to reinvent too many wheels. I want to do this because in my experience openni is only good when directly facing the camera.

I was wondering:

1. what would you recommend as a python toolkit for fragmenting and resizing image patches in python? opencv? mahotas?

2. where might I find good labeled training data for this type of project?

3. do you know of pakages already out there that do skeleton tracking with rotational invariance? do you know of any packages for studying body language using kinnect?

btw, I am not doing this for the kaggle competition that's currently happening.