Uncertainty estimation in a vision-based tracking system

Vision-based tracking is concerned with the recovery of position and orientation data of moving objects based on visual input provided by one or more cameras. This paper describes a framework to handle geometric parameter uncertainties within a monocular outside-in vision-based tracking application. We present a sensor model - the stochastic camera - that is capable to take parameter calibration uncertainties into consideration even under real-time requirements. The feasibility of the proposed method is shown in closed-loop tracking experiments.