An Approach for Target Tracking using Log-polar Images

Active vision brings important advantages for physically embodied artificial agents interacting with their environment. Gaze control is one of the important issues in active vision. In this paper, we address one subproblem of gaze control, namely, gaze stabilization, which appears when visually tracking a moving object is required. One approach to tackle this is by solving a motion estimation problem. On the other hand, foveal sensing is known to play an important role within active vision. Log-polar imaging is a biologically motivated foveal model with important benefits for tasks such as tracking. Therefore, here we propose an adaptation of a motion estimation algorithm, initially developed for cartesian images, to log-polar images. Experiments are included to illustrate the application of the approach to estimate the motion of a target in real image sequences, as well as to show how motion estimates can be used to drive a pan-tilt head, which conveys some benefits over the mere passive tracking approach.