A research paper (PDF format) from the CMU Robotics Institute describes a new approach to the problems of tracking and structure from motion (SFM) in robot vision. In previous systems tracking was done as a first step and the stabalized image was passed to a second step in which the structure of features was extracted. This has been a problematic approach in the real world because the tracking layer can't compensate for occlusions or abrupt camera motion. In his paper, Peng Chang proposes combining SFM with tracking. Having information about the structures and content of the view allow the tracking layer to reason about sudden changes in an intelligent way. The paper includes lots of example photos from a 360 degree view vision system on a test robot.