Featureless visual tracking based on non-vector space control theory

Abstract

© IFAC. This paper proposes a featureless visual target tracking approach based on non-vector space control theory. By considering the image as a set with pixels as its elements, the visual tracking problem could be treated as the mutation control between an initial image set and a prescribed target image set, then the motion of the robot can be reflected in the dynamic of the image set. Based on mutation analysis over sets, a shape functional describing the difference between two dynamic sets is defined, the directional derivative of this shape functional is derived and a Lyapunov function is constructed to design a controller to make an initial image set to track a moving goal image set, thereby steering the robot to follow the motion of the target. A 2-dimensional translation motion case is employed as an example to illustrate the feasibility of the proposed approach.