Due to the growing problem of space debris, research in the area of on-orbit servicing has intensified over the last years. Part of such missions is a controlled approach of a target object. To support control algorithms or human teleoperators in such a task, it can be helpful to detect the pose as well as the geometry of the object.
As an alternative to expensive and complicated LIDAR systems, RGB-D sensors could be an option. They allow to perform pose estimation and 3D reconstruction tasks by recording a color and a depth frame simultaneously. Applicable algorithms have advanced such that they can be run on a consumer desktop CPU in real-time.
In this thesis, current state-of-the-art algorithms have been selected for the RACOON hardware-in-the-loop-simulator to make use of a Kinect v2 camera that has been installed on a chaser satellite to estimate the pose of an unknown, uncooperative target satellite and create a 3D reconstruction. In comparison with traditional, feature-based methods, dense pose estimation and 3D reconstruction algorithms are used.
For this purpose, a software framework was developed, with which these and other algorithms and sensor types can be easily compared. By employing a flexible pipeline structer as well as wrapper classes with clearly defined interfaces for algorithms and data sources, easy extensibility and versatility are ensured.
With this thesis, a first step should be taken towards a comprehensive solution for pose estimation and 3D reconstruction in the context of teleoperated close-range navigation in orbit. Results of preliminary qualitative reconstruction tests show that the chosen algorithms are generally usable in the given simulation environment, even though extended functionalities for dift reduction are necessary to make them applicable for a teleoperation scenario.
«
Due to the growing problem of space debris, research in the area of on-orbit servicing has intensified over the last years. Part of such missions is a controlled approach of a target object. To support control algorithms or human teleoperators in such a task, it can be helpful to detect the pose as well as the geometry of the object.
As an alternative to expensive and complicated LIDAR systems, RGB-D sensors could be an option. They allow to perform pose estimation and 3D reconstruction tasks b...
»