Visual tracking of mobile robots in formation

Abstract

This paper presents a vision-based architecture for mobile robot detection and tracking from single frames using off-the-shelf mobile cameras and fiducial markers. A dual unscented Kalman filter is developed to estimate relative position, bearing, heading angles, and leader's velocities. The final goal of this framework is to support the design of robust vision-based algorithms to solve multi-robot coordination problems by eliminating the need of inter-vehicle communication. Simulation results verify the validity of the described architecture.