A smart hybrid vision aided inertial navigation system approach for UAVs in a GNSS denied environment

Abstract

This paper presents a smart hybrid vision-aided inertial navigation system (VAINS) for enhancing the navigation accuracy of UAVs in GNSS-denied environments. The proposed system consists of GNSS, INS, monocular visual odometry (VO) based on optical flow and regression trees, a Gaussian process regression (GPR), and extended Kalman filter (EKF) for data fusion. Although a variety of monocular VO-based on photogrammetric, structure from motion, and machine learning approaches have been proposed to assist the navigation process, the accuracy of 3-D positioning using these techniques is still affected by some factors such as the lack of the observed features, incorrect matches, and the accumulated positioning drift errors. Therefore, the development of a reliable navigation system that can handle such challenges is necessary. In this system, the GPR algorithm is deployed to model these errors when the GNSS signal is available and to correct them when the GNSS signal is lost.