Visual tracking for mobile robot pursuit

Abstract

This research is part of a broader effort to develop a supervisory control system for small robot navigation. Previous research and development focused on a one-touch, point-and-go navigation control system using visual homing. In the current research, we have begun to investigate visual tracking methods to extend supervisory control to tasks involving tracking and pursuit of a moving object. Ground-to-ground tracking of arbitrary targets in natural and damaged environments is challenging. Automatic tracking is expected to fail due to line-of-sight obstruction, lighting gradients, rapid changes in perspective and orientation, etc. In supervisory control, the automatic tracker needs able to alert the operator when it is at risk of losing track or when it may have already lost track, and do so with a low false alarm rate. The focus of the current research is on detecting tracking failure during pursuit. We are attempting to develop approaches to detecting failure that can integrate different low-level tracking algorithms. In this paper, we demonstrate stereo vision methods for pursuit tracking and examine several indicators of track loss in field experiments with a variety of moving targets in natural environment.