A pan/tilt/zoom tracking system for unmanned ground vehicles

Abstract

This thesis reports on the development of a pan/tilt/zoom camera tracking system capable of estimating the position of one moving vehicle relative to another, enabling robotic leader/follower behaviour. The system is composed of a number of sub-components including colour and SIFT based recognition, Kalman Filtering, and Linear Quadratic Gaussian control. A number of unique contributions are presented, including the ability to be trained at run-time and a novel zoom algorithm which balances the needs of computer vision with robust control. The complete system was demonstrated in a leader/follower scenario on the Defence R&D Canada experimental proving ground over a 1.75 km gravel road.