I'm working on a research project where we use Kinect (+ accelerometers) to detect one person's activities. We're using the skeleton provided by the SDK and it

I'm working on a research project where we use Kinect (+ accelerometers) to detect one person's activities. We're using the skeleton provided by the SDK and it doesn't work very well when the person is not facing the camera (which is understandable since it was originally designed for games). Also, the SDK skeleton doesn't seem to have any idea about morphological constraints (e.g. impossible arm rotations). Do you plan/is it possible to improve the skeleton tracking part ?