Large scale image aided navigation

Abstract

Currently, most airborne platforms use an Inertial Measurement Unit (IMU) to provide estimates of aircraft position, velocity, and attitude. Errors in the IMU result in a position solution that drifts over time. In order to counteract IMU drift, platforms typically integrate measurements from Global Navigation Satellite Systems (GNSS). However, many military and civilian scenarios encounter situations where GNSS may be unavailable due to intentional jamming, or when operating in challenging environments. Many solutions have been proposed that use images from an airborne camera to match features against a reference image to compute global position. These techniques require a relatively accurate prior estimate of position to converge. This leads to a situation where platforms being denied GNSS pass a threshold of position error due to IMU drift, from which current image-aided navigation system cannot recover. This paper presents a novel localization algorithm inspired by fusing state of the art Content Based Image Retrieval techniques with nonparametric Bayesian estimation strategies. We outline the framework for performing a search of a large-scale image database in order to provide a rough position estimate to an airborne platform. These techniques are analogous to solving the “kidnapped-robot” problem, applied to airborne platforms. This framework is evaluated using actual data collected from an airborne camera and navigation system.