Tracking with Hybrid-Drive Active vision

Abstract

The work in this paper demonstrates eective tracking in an everyday environment with hybrid-drive active vision. We base our approach on the assumption that to eciently perceive an unknown, unconstrained, potentially evolving environment, the observing system must be able to process and act on gathered visual information as it happens in real-time. Moreover, we endow the robot's tracking system with two behavioural subgroups, saccade and smooth pursuit and show that saccade time and motion smoothness can be optimised through the use of a single trapezoidal prole motion (TPM) algorithm. In this way, the simplicity and compactness of the controller's design is improved. To demonstrate this, we present a 4 degree of freedom binocular active vision platform, nicknamed HyDrA, that uses cable-drive technology to minimise mass-in-motion and hence improve performance. By combining template matching and TPM control, we show that HyDrA can detect and track the motion of a...