Directional Statistics for 3D Model-Based UAV Tracking

Abstract

We present a vision system based on a monocular camera to track the 3D position and orientation of an Unmanned Aerial Vehicle (UAV) during the landing process aboard a ship. The proposed method uses a 3D model-based approach based on a Particle Filter (PF) with proposal distributions given by an Unscented Kalman Filter (UKF) for the translational motion and filters based on directional statistics for the rotational motion. Our main contributions are (i) the development of a novel 3D model-based tracking architecture based on directional statistics that can be easily adapted to other tracking problems, and (ii) the development of the Unscented Bingham-Gauss Filter (UBiGaF) for rotation estimation. We show the advantages of using directional statistics based filters on 3D model-based tracking in a series of quantitative tests in a challenging simulation scenario with real video data. The obtained position and angular error are compatible with the automatic landing system requirements when using directional statistics. We obtain lower error when using the UBiGaF scheme for the vast majority of the tested combinations.