Combined Vision – Inertial Navigation with Improved Outlier Robustness

Abstract

This paper describes a loosely coupled approach for the improvement of state estimation in autonomous inertial navigation, using image-based relative motion estimation for augmentation. The augmentation system uses a recently proposed pose estimation technique based on a Entropy-Like cost function, which was proven to be robust to the presence of noise and outliers in the visual features. Experimental evidence of its performance is given and compared to a state-of-the-art algorithm. Vision-inertial integrated navigation is achieved using an Indirect Kalman Navigation Filter in the framework of stochastic cloning, and the proposed robust relative pose estimation technique is used to feed a relative position fix to the navigation filter. Simulation and Experimental results are presented and compared with the results obtained via the classical RANSAC – based Direct Linear Transform approach.