Motion Uncertainty and Field of View
Anat Levin, Richard Szeliski · 2006
One of the important image and video registration goals is the accurate motion and structure estimation. On the other hand, a good motion estimation is also an impor-tant requirement for most mosaicing and novel view generation techniques. While it has been well known for a while that narrow field-of-view cameras have a hard time distinguishing between certain kinds of rotations and translations, it has been recently observed that using omni-directional cameras significantly decrease those ambiguities. In this paper we study the relationship between field of view and the amount of mo-tion uncertainty. To this goal, we derive an analytic formula for the Fisher information matrix involved in the motion estimation task. We use this to examine the coupling be-tween the different motion parameters and the relation between those uncertainties to the camera filed of view. In particular, we provide a formal proof to the previously ob-served phenomena, that for full 360o omni-directional cameras there is no correlation between rotation and translation parameters estimate.