On visual pursuit systems

Abstract

Visual pursuit systems are reviewed using a generalized form with the principal delay in the visual feedback pathway. The performance of several different control methods are compared through analysis of the tracking error. A high-performance active vision system, called TRICLOPS, is used to obtain experimental data on the tracking performance of the various methods, some of which had previously only been tested through simulation. Image processing delay is shown to be the primary limiting factor in performing high-speed tracking. Prediction is necessary to achieve high performance tracking, and the effectiveness of prediction depends on several factors, including the coordinate system used for motion modeling and the noise content of the target motion signal. It is demonstrated that, using relatively simple tracking algorithms which incorporate prediction with an advanced robotic device like TRICLOPS, it is now possible to outperform humans in the domain of visual pursuit of simple targets.