Event-driven 3D reconstruction of small celestial bodies in a simulated laboratory environment

This thesis investigates pose estimation algorithms for event cameras. Event cameras are innovative devices that detect motion faster than the update rate of traditional videos, and computer vision algorithms employ the new visual input to estimate trajectories in terrestrial or space applications. In particular, this work describes pose estimation in a simulated environment of deep space navigation, namely a spacecraft’s circular orbit around a comet. From the simulation, the code must detect features, track their evolution, and establish how the camera moves without human intervention. Two main procedures are implemented to achieve this goal, one based entirely on event cameras and the other combining traditional and event videos. Together with the results, a complete analysis of the algorithms is provided: starting from the raw data captured from the scene, this thesis will describe how computer vision determines the relative motion between the camera and the features and how it represents the scene in three-dimensional coordinates.