Omnidirectional imaging and optical flow

Catadioptric imaging sensors provide a 360° horizontal view which may be advantageous for visual surveillance and motion detection in robotic applications. Larger object surfaces can be perceived in a wide angle of view supplied by such an omnidirectional camera. The duration of object appearance in the field of view is also longer. These properties cause smooth variations of the image scenes and facilitate the calculation of the optical flow. Using such a flow field allows to detect ego-motion and other fundamental preliminaries of robot navigation. In this paper we review the motion detection algorithms, which may be appropriate for optical flow calculations on omnidirectional images.