Image Gradient-based Monocular Visual-Inertial Odometry

Abstract

This paper presents image gradient-based monocular visual-inertial odometry (VIO) algorithm, using image gradient measurements, robust to illumination change. We expand the measurements from the reprojected feature locations on the image coordinates to the corresponding image gradients. The iterated EKF and low-pass pyramid are adapted to reduce the linearization error in the multi-state constraint Kalman filter (MSCKF) measurement update process. We verify that our proposed algorithm outperforms both conventional indirect and direct MSCKF-based VIO algorithms by evaluating the pose estimation performance.