Vision-based Navigation for Autonomous Rendezvous with Non-cooperative Targets PDF Download

Author: Anthea Comellini
Publisher:
ISBN:
Category :
Languages : en
Pages : 0

Get Book Here

Book Description
The aim of this thesis is to propose a full vision-based solution to enable autonomousnavigation of a chaser spacecraft (S/C) during close-proximity operations in space rendezvous(RDV) with a non-cooperative target using a visible monocular camera.Autonomous rendezvous is a key capability to answer main challenges in space engineering,such as Active Debris Removal (ADR) and On-Orbit-Servicing (OOS). ADR aimsat removing the space debris, in low-Earth-orbit protected region, that are more likelyto lead to future collision and feed the Kessler syndrome, thus increasing the risk foroperative spacecrafts. OOS includes inspection, maintenance, repair, assembly, refuelingand life extension services to orbiting S/C or structures. During an autonomous RDVwith a non-cooperative target, i.e., a target that does not assist the chaser in acquisition,tracking and rendezvous operations, the chaser must estimate the target's state on-boardautonomously. Autonomous RDV operations require accurate, up-to-date measurementsof the relative pose (i.e., position and attitude) of the target, and the combination ofcamera sensors with tracking algorithms can provide a cost effective solution.The research has been divided into three main studies: the development of an algorithmenabling the initial pose acquisition (i.e., the determination of the pose without any priorknowledge of the pose of the target at the previous instants), the development of a recursivetracking algorithm (i.e., an algorithm which exploits the information about thestate of the target at the previous instant to compute the pose update at the currentinstant), and the development of a navigation filter integrating the measurements comingfrom different sensor and/or algorithms, with different rates and delays.For what concerns the pose acquisition phase, a novel detection algorithm has been developedto enable fast pose initialization. An approach is proposed to fully retrieve theobject's pose using a set of invariants and geometric moments (i.e., global features) computedusing the silhouette images of the target. Global features synthesize the content ofthe image in a vector of few descriptors which change values as a function of the targetrelative pose. A database of global features is pre-computed offline using the target geometricalmodel in order to cover all the solution space. At run-time, global features arecomputed on the current acquired image and compared with the database. Different setsof global features have been compared in order to select the more performing, resultingin a robust detection algorithm having a low computational load.Once an initial estimate of the pose is acquired, a recursive tracking algorithm is initialized.The algorithm relies on the detection and matching of the observed silhouettecontours with the 3D geometric model of the target, which is projected into the imageframe using the estimated pose at the previous instant. Then, the summation of the distances between each projected model points and the matched image points is written as a non-linear function of the unknown pose parameters. The minimization of this costfunction enables the estimation of the pose at the current instant. This algorithm providesfast and very accurate measurements of the relative pose of the target. However,as other recursive trackers, it is prone to divergence. Thus, the detection algorithm isrun in parallel to the tacker in order to provide corrected measurements in case of trackerdivergences. The measurements are then integrated into the chaser navigation filter to provide anoptimal and robust estimate. Vision-based navigation algorithms provide only pose measurements.