Towards implementation of far background tracker for vision-based robot navigation

Abstract

Visual tracking of stationary points positioned at a large distance from moving robot provides a good basis for determining robot orientation and it may speed-up operation of structure from motion and navigation algorithms. This paper considers an implementation of the tracker that builds a far background model by assigning each visually tracked point a degree of membership that express an extent to which an apparent motion of corresponding image patch is in accordance to motion of projections of points at infinity. An experimental implementation of the tracker is described together with illustrative examples of its operation.