A Monocular Vision-Based Algorithm for Spacecraft Pose Estimation

Ziyi He; Feng Zhan; Jian Chen; Ruichen Xi

doi:10.1016/j.ifacol.2025.11.158

A Monocular Vision-Based Algorithm for Spacecraft Pose Estimation

Ziyi He^*
, Feng Zhan
, Jian Chen^*
, Ruichen Xi^*

^*Corresponding author for this work

School of Astronautics

Research output: Contribution to journal › Conference article › peer-review

Abstract

A monocular vision-based algorithm is developed for spacecraft pose estimation aiming to enhance the navigation accuracy and efficiency. Firstly, a new multi-task learning network is proposed, using object detection, six-degree of freedom (6-DoF) estimation, keypoint regression and instance segmentation. Subsequently, an end-to-end Perspective-n-Point (PnP) approach is designed through differentiable iterative optimization, using adaptive multiple importance sampling (AMIS), predicting the discrete probability distribution of 6-DoF on SE(3). Experimental results on a public dataset, along with comparisons to State-Of-The-Art (SOTA) algorithms demonstrate the effectiveness of the algorithm.

Original language	English
Pages (from-to)	250-255
Number of pages	6
Journal	IFAC-PapersOnLine
Volume	59
Issue number	20
DOIs	https://doi.org/10.1016/j.ifacol.2025.11.158
State	Published - 1 Aug 2025
Event	23th IFAC Symposium on Automatic Control in Aerospace, ACA 2025 - Harbin, China Duration: 2 Aug 2025 → 6 Aug 2025

Keywords

Adaptive multiple importance sampling
End-to-end PnP
Monocular vision navigation
Multi-task learning
Pose estimation

Access to Document

10.1016/j.ifacol.2025.11.158

Cite this

@article{361b1d5999984d308a144ea73b363443,

title = "A Monocular Vision-Based Algorithm for Spacecraft Pose Estimation",

abstract = "A monocular vision-based algorithm is developed for spacecraft pose estimation aiming to enhance the navigation accuracy and efficiency. Firstly, a new multi-task learning network is proposed, using object detection, six-degree of freedom (6-DoF) estimation, keypoint regression and instance segmentation. Subsequently, an end-to-end Perspective-n-Point (PnP) approach is designed through differentiable iterative optimization, using adaptive multiple importance sampling (AMIS), predicting the discrete probability distribution of 6-DoF on SE(3). Experimental results on a public dataset, along with comparisons to State-Of-The-Art (SOTA) algorithms demonstrate the effectiveness of the algorithm.",

keywords = "Adaptive multiple importance sampling, End-to-end PnP, Monocular vision navigation, Multi-task learning, Pose estimation",

author = "Ziyi He and Feng Zhan and Jian Chen and Ruichen Xi",

note = "Publisher Copyright: Copyright {\textcopyright} 2025 The Authors.; 23th IFAC Symposium on Automatic Control in Aerospace, ACA 2025 ; Conference date: 02-08-2025 Through 06-08-2025",

year = "2025",

month = aug,

day = "1",

doi = "10.1016/j.ifacol.2025.11.158",

language = "英语",

volume = "59",

pages = "250--255",

journal = "IFAC-PapersOnLine",

issn = "2405-8971",

publisher = "Elsevier B.V.",

number = "20",

}

TY - JOUR

T1 - A Monocular Vision-Based Algorithm for Spacecraft Pose Estimation

AU - He, Ziyi

AU - Zhan, Feng

AU - Chen, Jian

AU - Xi, Ruichen

PY - 2025/8/1

Y1 - 2025/8/1

N2 - A monocular vision-based algorithm is developed for spacecraft pose estimation aiming to enhance the navigation accuracy and efficiency. Firstly, a new multi-task learning network is proposed, using object detection, six-degree of freedom (6-DoF) estimation, keypoint regression and instance segmentation. Subsequently, an end-to-end Perspective-n-Point (PnP) approach is designed through differentiable iterative optimization, using adaptive multiple importance sampling (AMIS), predicting the discrete probability distribution of 6-DoF on SE(3). Experimental results on a public dataset, along with comparisons to State-Of-The-Art (SOTA) algorithms demonstrate the effectiveness of the algorithm.

AB - A monocular vision-based algorithm is developed for spacecraft pose estimation aiming to enhance the navigation accuracy and efficiency. Firstly, a new multi-task learning network is proposed, using object detection, six-degree of freedom (6-DoF) estimation, keypoint regression and instance segmentation. Subsequently, an end-to-end Perspective-n-Point (PnP) approach is designed through differentiable iterative optimization, using adaptive multiple importance sampling (AMIS), predicting the discrete probability distribution of 6-DoF on SE(3). Experimental results on a public dataset, along with comparisons to State-Of-The-Art (SOTA) algorithms demonstrate the effectiveness of the algorithm.

KW - Adaptive multiple importance sampling

KW - End-to-end PnP

KW - Monocular vision navigation

KW - Multi-task learning

KW - Pose estimation

UR - https://www.scopus.com/pages/publications/105025983650

U2 - 10.1016/j.ifacol.2025.11.158

DO - 10.1016/j.ifacol.2025.11.158

M3 - 会议文章

AN - SCOPUS:105025983650

SN - 2405-8971

VL - 59

SP - 250

EP - 255

JO - IFAC-PapersOnLine

JF - IFAC-PapersOnLine

IS - 20

T2 - 23th IFAC Symposium on Automatic Control in Aerospace, ACA 2025

Y2 - 2 August 2025 through 6 August 2025

ER -

A Monocular Vision-Based Algorithm for Spacecraft Pose Estimation

Abstract

Keywords

Access to Document

Other files and links

Fingerprint

Cite this