Evaluation of tightly- and loosely-coupled approaches in CNN-based pose estimation systems for uncooperative spacecraft

Cassinis, Lorenzo Pasqualetto; Fonod, Robert; Gill, Eberhard; Ahrns, Ingo; Gil-Fernández, Jesús

doi:10.1016/j.actaastro.2021.01.035

Cited by 37 publications

(5 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…After SPEC, published DNN-based work has seldom considered actual rendezvous trajectories [34], continuing to focus instead on individual greyscale images of SPEED [35][36][37]. In either case, the proposed strategies consist in using a CNN for keypoint detection for use with PnP.…”

Section: B Learning-based Methodsmentioning

confidence: 99%

ChiNet: Deep Recurrent Convolutional Learning for Multimodal Spacecraft Pose Estimation

Rondao

Aouf

Richardson

2022

IEEE Trans. Aerosp. Electron. Syst.

View full text Add to dashboard Cite

This paper presents an innovative deep learning pipeline which estimates the relative pose of a spacecraft by incorporating the temporal information from a rendezvous sequence. It leverages the performance of long short-term memory (LSTM) units in modelling sequences of data for the processing of features extracted by a convolutional neural network (CNN) backbone. Three distinct training strategies, which follow a coarse-to-fine funnelled approach, are combined to facilitate feature learning and improve end-to-end pose estimation by regression. The capability of CNNs to autonomously ascertain feature representations from images is exploited to fuse thermal infrared data with electro-optical red-green-blue (RGB) inputs, thus mitigating the effects of artifacts from imaging space objects in the visible wavelength. Each contribution of the proposed framework, dubbed ChiNet, is demonstrated on a synthetic dataset, and the complete pipeline is validated on experimental data.

show abstract

Section: B Learning-based Methodsmentioning

confidence: 99%

ChiNet: Deep Recurrent Convolutional Learning for Multimodal Spacecraft Pose Estimation

Rondao

Aouf

Richardson

2022

IEEE Trans. Aerosp. Electron. Syst.

View full text Add to dashboard Cite

show abstract

“…In particular, this latter approach struggles to be compliant with on-board resources due to database storage and its computationally heavy search. This problem has been tackled in recent years by applying Template Matching datasets to Artificial Intelligence approaches through the use of Convolutional Neural Networks [20,[24][25][26]. Alternatively, pose can be determined by taking the target geometrical characteristics into account during the design process [27].…”

Section: Literature Reviewmentioning

confidence: 99%

Relative Navigation Strategy About Unknown and Uncooperative Targets

Maestrini

Luca

Lizia

2023

Journal of Guidance, Control, and Dynamics

View full text Add to dashboard Cite

In recent years, space debris has become a threat for satellites operating in low Earth orbit. Even by applying mitigation guidelines, their number will still increase over the course of the century. As a consequence, active debris removal missions and on-orbit servicing missions have gained momentum at both academic and industrial level. The crucial step in both scenarios is the capability of navigating in the neighborhood of a target resident space object. This problem has been tackled many times in literature with varying level of cooperativeness of the target required. While several techniques are available when the target is cooperative or its shape is known, no approach is mature enough to deal with uncooperative and unknown targets. This paper proposes a hybrid method to tackle this issue called Coarse Model-Based Relative Navigation (CoMBiNa). The main idea of this algorithm is to split the mission into two phases. During the first phase, the algorithm constructs a coarse model of the target. In the second phase, this coarse model is used as a reference for a relative navigation technique, effectively shifting the focus toward state and inertia estimation. In addition, this paper proposes a strategy to leverage the structure of the selected navigation method to detect and reject outliers. To conclude, CoMBiNa is tested on a simulated environment to highlight its benefits and its shortcomings, while also assessing its applicability on a limited-resource single-board computer.

show abstract

“…The main reason for an emerging interest in CNNs for features extraction lies in the capability of their convolutional layers to extract high-level features of objects with improved robustness against image noise and illumination conditions as compared to standard IP algorithms [30]. As shown in Fig.…”

Section: Convolutional Neural Networkmentioning

confidence: 99%

“…Building on the authors' previous findings [17,30] and inspired by the promising texture randomization results presented in earlier works [28], the main objective of this paper is to investigate the impact of training data augmentation on the CNN performance on representative space imagery generated on-ground. In order to do so, special focus is put on the recreation of a dedicated calibration pipeline to validate the proposed pose estimation system on representative rendezvous scenarios.…”

Section: Introductionmentioning

confidence: 99%