In the last decades, space debris has gained great interest in the research community. Nowadays satellites are used for different applications such as Earth observation missions, surveillance, and telecommunications. This have led to an intensive exploitation of the space around the Earth, and in particular LEO orbits. In 2013 NASA has reported that more than 21,000 space debris have been tracked. Due to the recent growth of interest in space activity, the situation will get worse. For this reason, the attention has been shifted from short-term to long-term actions, i.e. the design of the mission of On-Orbit Servicing or Active Debris Removal. In particular, the research has moved to the direction of the design of an algorithm capable to navigate around a non-cooperative and unknown target using a monocular camera. To reach this objective different new algorithms and the ones already employed on-ground for navigation of robots have been studied. Visual Odometry is one of them and has attracted great interest. The focus of this work is the design of a monocular Visual Odometry algorithm. This kind of algorithm has been employed in on-ground mission but never proposed for on-orbit missions. The CAD model of an old mission of NASA, Clementine, has been used to simulate the target. The chaser is put on a closed relative orbit around the target that is considered in a circular orbit around the Earth. A simulation of the relative orbit has been imported on Blender and the rendering action, as the movement of the camera, have been automatized thanks to the API of the tool. Then the photos and information about the relative pose of the chaser with respect of the target have been imported to Matlab. The algorithm proposed tries to estimate the relative pose up to the knowledge of the initial pose of the chaser with respect to the target. By doing it, the reconstruction of the target body is possible given the knowledge of the relative position of the cameras. The first part of the thesis will cover the theoretical aspects of the problem. Then the motivations behind the customization of the tool Blender and the choices of the specific relative orbit will be given. After the detailed description of the algorithm structure, the results are shown. The result of each step is shown and analyzed. Also, a validation of the essential camera matrix computation has been made.
Nell’ultimo decennio gli space debris hanno suscitato grande interesse nella comunità scientifica. Oggi i satelliti sono utilizzati per diverse applicazioni, come missioni di osservazione della Terra, la sorveglianza e le telecomunicazioni. Questo ha portato a uno sfruttamento intensivo dello spazio antorno alla Terra, in particolare delle orbite LEO. Nel 2013 la NASA ha riferito che più di 21.000 detriti spaziali sono stati rintracciati. A causa del recente aumento di interesse nelle missioni spaziali, la situazione è destinata a peggiorare. Per questo motivo l’attenzione è stata spostata da azioni a breve termine a quelle a lungo termine, ovvero, alla progettazione di missioni di On-Orbit Servicing o di Active Debris Removal. In particolare, la ricerca si è spostata nella direzione della progettazione di un algoritmo in grado di navigare attorno a un target sconosciuto e non cooperativo utilizzando una telecamera monoculare. Per raggiungere questo obiettivo sono stati studiati nuovi algoritmi e modificati quelli già impiegati a terra per la navigazione di robot. Il Visual Odometry è uno di questi e ha suscitato grande interesse all’intero della comunità scientifica. L’obiettivo di questa tesi è la progettazione di un algoritmo di Monocular Visual Odometry. Questo tipo di algoritmo è stato impiegato in missioni a terra, ma non è mai stato proposto per missioni in orbita. Il modello di una vecchia missione della NASA, Clementine, è stato usato per simulare il target. Il chaser viene messo su un’orbita relativa chiusa attorno al target che viene considerato in un’orbita circolare attorno alla Terra. Una simulazione dell’orbita relativa è stata importata con Blender e il rendering e il movimento della telecamera sono stati automatizzati grazie all’API del tool. Poi le foto e le informazioni sulla posa relativa del chaser rispetto al bersaglio sono state importate su Matlab. L’algoritmo proposto cerca di stimare la posa relativa, conoscendo la posa iniziale dell’chaser rispetto al target. In questo modo la ricostruzione del modello del target è possibile, data la conoscenza della posizione relativa delle telecamere. La prima parte della tesi tratterà gli aspetti teorici del problema. Poi i motivi che stanno alla base della castomizzazione del tool Blender e la scelta di un’orbita relativa specifica verranno discusse. Dopo la descrizione dettagliata della struttura dell’algoritmo vengono mostrati i risultati. Il risultato di ogni fase viene mostrato e analizzato. È stata inoltre effettuata una validazione del calcolo della essential matrix.
Monocular Visual Odometry for Navigation around a Non-Cooperative and Unknown Target
CALDERAIO, MANFREDI
2021/2022
Abstract
In the last decades, space debris has gained great interest in the research community. Nowadays satellites are used for different applications such as Earth observation missions, surveillance, and telecommunications. This have led to an intensive exploitation of the space around the Earth, and in particular LEO orbits. In 2013 NASA has reported that more than 21,000 space debris have been tracked. Due to the recent growth of interest in space activity, the situation will get worse. For this reason, the attention has been shifted from short-term to long-term actions, i.e. the design of the mission of On-Orbit Servicing or Active Debris Removal. In particular, the research has moved to the direction of the design of an algorithm capable to navigate around a non-cooperative and unknown target using a monocular camera. To reach this objective different new algorithms and the ones already employed on-ground for navigation of robots have been studied. Visual Odometry is one of them and has attracted great interest. The focus of this work is the design of a monocular Visual Odometry algorithm. This kind of algorithm has been employed in on-ground mission but never proposed for on-orbit missions. The CAD model of an old mission of NASA, Clementine, has been used to simulate the target. The chaser is put on a closed relative orbit around the target that is considered in a circular orbit around the Earth. A simulation of the relative orbit has been imported on Blender and the rendering action, as the movement of the camera, have been automatized thanks to the API of the tool. Then the photos and information about the relative pose of the chaser with respect of the target have been imported to Matlab. The algorithm proposed tries to estimate the relative pose up to the knowledge of the initial pose of the chaser with respect to the target. By doing it, the reconstruction of the target body is possible given the knowledge of the relative position of the cameras. The first part of the thesis will cover the theoretical aspects of the problem. Then the motivations behind the customization of the tool Blender and the choices of the specific relative orbit will be given. After the detailed description of the algorithm structure, the results are shown. The result of each step is shown and analyzed. Also, a validation of the essential camera matrix computation has been made.| File | Dimensione | Formato | |
|---|---|---|---|
|
Tesi_Manfredi_Calderaio_Final.pdf
Open Access dal 30/11/2023
Dimensione
5.53 MB
Formato
Adobe PDF
|
5.53 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/198286