Body-Centered Dynamically-Tuned Error-State Extended Kalman Filter for Visual Inertial Odometry in GNSS-Denied Environments

Abstract

The use of visual odometry in autonomous land vehicle and unmanned ground vehicle (UGV) applications has shown to be promising in providing robust and accurate positioning and navigation in GNSS-denied environments. While GNSS is the most accurate and reliable positioning source, it cannot be received indoors. Furthermore, low-cost micro-electro-mechanical system (MEMS)-based inertial sensors are used for positioning but can result in rapid drift during prolonged GNSS outages. To address these limitations, this paper proposes an innovative method of integrating MEMS-based inertial sensors with a vision-based navigation system. The vision-based navigation relies on a visual odometry pipeline modified to replace the feature-matching step with Lucas-Kanade optical flow. While the inertial-based system is derived from the three-dimensional reduced inertial sensor system (3D-RISS) and modified for a bodycentered update. The fusion of both systems is based on an Extended Kalman filter (EKF) to estimate the position and heading information and correct for inertial sensor errors. To assess the effectiveness of the proposed method, real indoor and road test trajectories were conducted using both a Husky A200 UGV and a sensor-equipped minivan. The results show that the body-centered inertial system, aided with reliable vision updates, provides a reliable positioning solution for an extended duration in GNSS-denied environments.