Global Localization of Ground Vehicles Using Self-Describing Fiducials Coupled with IMU Data

Abstract

A key aspect of providing safe, reliable navigation for autonomous vehicles is accurate localization. This is often accomplished with the use of GPS in conjunction with odometry provided by other measurement systems. However, in many cases GPS is not available, or its accuracy is severely degraded allowing odometry error to propagate to unacceptable levels. Much work that addresses this issue either uses LIDAR, which is too expensive, bulky, and heavy for some applications, or computer vision, which often requires too much computation power for many of the same applications. Self-describing fidu-cials, fiducials which provide their own location information, can be a lower-cost, and more usable method of providing global location information to an autonomous ground vehicle. To this end this work details a low-cost ground vehicle localization method that uses inertial odometry and self-describing visual fiducials, combined through an indirect extended Kalman filter, for use in GPS-denied or degraded environments. Additionally, the sensitivity of the localization to fiducial density and IMU grade are analyzed.