Vision-aided Inertial System for Near-real-time Positioning and Navigation of Unmanned Ground Survey Vehicles in GNSS-denied Environments
Loading...
Date
Authors
Kaoud Abdelaziz, Shaza
Keyword
Inertial Navigation , Visual Odometry , GNSS-denied Navigation , Indirect Kalman Filter , Sensor Fusion
Abstract
Positioning and navigation of unmanned ground vehicles (UGVs) in GNSS-denied environments is crucial to geological surveys required for subterranean and extraterrestrial exploration and mapping. While GNSS is the most accurate and reliable source of positioning service when operating in open sky environments, many field sites of geological interest for exploration and mining cannot receive GNSS signals. The use of robotic platforms helped improve data resolution and data collection speed compared to previous survey methods. The technological advances and the low cost of micro-electro-mechanical system (MEMS)-based inertial sensors (accelerometers and gyroscopes) enabled their use in positioning and navigation for many years. However, during prolonged surveys, the inertial-based positioning solution may become prone to rapid drift resulting in substantial position errors. This research explores integrating MEMS-based inertial sensors with a vision-based navigation system. The work completed in this thesis proposes a new method for aiding the inertial-based navigation solution with the vision-based navigation solution. Each of the subsystems is modified in order to optimize the navigation solution. The vision-based navigation relies on a visual odometry pipeline modified to replace the feature matching step with Lucas-Kanade optical flow. Simultaneously, the inertial-based subsystem is derived from the three-dimensional reduced inertial sensor system (3D-RISS), modified for a body-centred update. A unique body-centred approach relying on projecting position increments through time is used to minimize the errors previously occurring due to the correlation of position increment and heading angle. The fusion of both systems is based on a Kalman filter to estimate the position, heading and inertial sensor errors. The developed method was tested using a Husky A200 UGV. The field tests utilized the Robotic Operating System (ROS) for recording data from sensors and synchronization. The test trajectories were performed indoors at the Royal Military College of Canada in Kingston, Ontario and in the Brockville Railway Tunnel in Brockville, Ontario. This thesis presents illuminating findings of aiding a body-centred inertial system with only reliable vision updates, resulting in a reliable positioning and navigation solution operating for an extended duration in GNSS-denied environments.