Vision-based Integrated Pose Estimation of UAS and Moving Platforms

This work uses visual-inertial methods to estimate the integrated pose of an Unmanned Aerial System (UAS) and a non-cooperative moving platform in a GPS-denied environment. A single integrated navigation filter comprising an Extended Kalman Filter (EKF) is used to estimate the states of the UAS and platform. Stereo vision is used to compute the depth of feature points used in the estimation process. A generic process model is adopted for the platform that allows the pose estimation algorithm to be agnostic to the type of platform. The f ilter is also capable of estimating the shape of the platform. The algorithm can be used for applications such as autonomous ship deck landing, station keeping over a ship, tracking and following a moving vehicle, and obstacle avoidance. While tests are performed in a GPS-denied environment, the system can use GPS data, when available, to improve the filter performance, thus allowing both indoor and outdoor operations. Flight test results of the integrated pose estimation algorithm in an indoor setting are presented with the platform moving over a level terrain.