Vision-aided terrain referenced navigation for unmanned aerial vehicles using ground features

Abstract

A vision-aided terrain referenced navigation (VATRN) approach is addressed for autonomous navigation of unmanned aerial vehicles (UAVs) under GPS-denied conditions. A typical terrain referenced navigation (TRN) algorithm blends inertial navigation data with measured terrain information to estimate vehicle's position. In this paper, a low-cost inertial navigation system (INS) for UAVs is supplemented with a monocular vision-aided navigation system and terrain height measurements. A point mass filter based on Bayesian estimation is employed as a TRN algorithm. Homograpies are established to estimate the vehicle's relative translational motion using ground features with simple assumptions. And the error analysis in homography estimation is explored to estimate the error covariance matrix associated with the visual odometry data. The estimated error covariance is delivered to the TRN algorithm for robust estimation. Furthermore, multiple ground features tracked by image observations are utilized as multiple height measurements to improve the performance of the VATRN algorithm.