Title:State Estimation for Vision-based Localization under Uncertain Conditions

Abstract:Vision based localization is a popular approach to carry out manoeuvres particularly in GPS-restricted indoor environments, because vision can complement other activities performed by the robot. The objective is to estimate the current location with respect to a known location by matching the bearings. The problem is challenging as the known location information is in terms of the bearings of landmarks extracted from an image. We address the problem under more challenging scenario when landmarks are semi-static. In this work, an observer formulation is presented which enables to incorporate the effect of change in landmark position as parameters. The efficacy of two estimators: Augmented Extended Kalman Filter (A-EKF) and a Proportional-Integral EKF (PI-EKF) is tested under the cases where there are changes in some of the landmark positions. Morever, it is likely that not all landmarks are visible to the robot at all instants of time. A multi-rate estimation framework is proposed to mitigate this issue. Observability analysis is carried out to arrive upon a minimum number of landmarks required for such a formulation. Simulation studies are presented to test the efficacy of the formulations.