Fast Relative Pose Estimation using Relative Depth
(2024) 11th International Conference on 3D Vision, 3DV 2024 In Proceedings - 2024 International Conference on 3D Vision, 3DV 2024 p.873-881- Abstract
In this paper, we revisit the problem of estimating the relative pose from a sparse set of point-correspondences. For each point-correspondence we also estimate the relative depth, i.e. the relative distance to the scene point in the two images. This yields an additional constraint, allowing us to use fewer matches in RANSAC to generate the pose candidates. In the paper we propose two novel minimal solvers: one for general motion and one for the case of known vertical direction. To obtain the relative depth estimates, we explore using scale estimates obtained from a keypoint detector as well as a neural network that directly predicts the relative depth for a pair of patches. We show in experiments that while our estimates are more noisy... (More)
In this paper, we revisit the problem of estimating the relative pose from a sparse set of point-correspondences. For each point-correspondence we also estimate the relative depth, i.e. the relative distance to the scene point in the two images. This yields an additional constraint, allowing us to use fewer matches in RANSAC to generate the pose candidates. In the paper we propose two novel minimal solvers: one for general motion and one for the case of known vertical direction. To obtain the relative depth estimates, we explore using scale estimates obtained from a keypoint detector as well as a neural network that directly predicts the relative depth for a pair of patches. We show in experiments that while our estimates are more noisy compared to the purely point-based solvers, the smaller sample size leads to a significantly reduced runtime in settings with high outlier ratios.
(Less)
- author
- Astermark, Jonathan
LU
; Ding, Yaqing
LU
; Larsson, Viktor
LU
and Heyden, Anders
LU
- organization
-
- Computer Vision and Machine Learning (research group)
- LU Profile Area: Natural and Artificial Cognition
- LTH Profile Area: AI and Digitalization
- Mathematics (Faculty of Engineering)
- Mathematical Imaging Group (research group)
- LU Profile Area: Proactive Ageing
- LTH Profile Area: Engineering Health
- ELLIIT: the Linköping-Lund initiative on IT and mobile communication
- eSSENCE: The e-Science Collaboration
- publishing date
- 2024
- type
- Chapter in Book/Report/Conference proceeding
- publication status
- published
- subject
- keywords
- epipolar geometry, minimal solver, RANSAC, relative depth, Relative pose, SIFT
- host publication
- Proceedings - 2024 International Conference on 3D Vision, 3DV 2024
- series title
- Proceedings - 2024 International Conference on 3D Vision, 3DV 2024
- pages
- 9 pages
- publisher
- IEEE - Institute of Electrical and Electronics Engineers Inc.
- conference name
- 11th International Conference on 3D Vision, 3DV 2024
- conference location
- Davos, Switzerland
- conference dates
- 2024-03-18 - 2024-03-21
- external identifiers
-
- scopus:85196743754
- ISBN
- 9798350362459
- DOI
- 10.1109/3DV62453.2024.00053
- language
- English
- LU publication?
- yes
- id
- dba2f06b-76f5-43d2-89c2-b7fa7f39a590
- date added to LUP
- 2024-08-29 15:41:59
- date last changed
- 2024-08-29 15:43:02
@inproceedings{dba2f06b-76f5-43d2-89c2-b7fa7f39a590, abstract = {{<p>In this paper, we revisit the problem of estimating the relative pose from a sparse set of point-correspondences. For each point-correspondence we also estimate the relative depth, i.e. the relative distance to the scene point in the two images. This yields an additional constraint, allowing us to use fewer matches in RANSAC to generate the pose candidates. In the paper we propose two novel minimal solvers: one for general motion and one for the case of known vertical direction. To obtain the relative depth estimates, we explore using scale estimates obtained from a keypoint detector as well as a neural network that directly predicts the relative depth for a pair of patches. We show in experiments that while our estimates are more noisy compared to the purely point-based solvers, the smaller sample size leads to a significantly reduced runtime in settings with high outlier ratios.</p>}}, author = {{Astermark, Jonathan and Ding, Yaqing and Larsson, Viktor and Heyden, Anders}}, booktitle = {{Proceedings - 2024 International Conference on 3D Vision, 3DV 2024}}, isbn = {{9798350362459}}, keywords = {{epipolar geometry; minimal solver; RANSAC; relative depth; Relative pose; SIFT}}, language = {{eng}}, pages = {{873--881}}, publisher = {{IEEE - Institute of Electrical and Electronics Engineers Inc.}}, series = {{Proceedings - 2024 International Conference on 3D Vision, 3DV 2024}}, title = {{Fast Relative Pose Estimation using Relative Depth}}, url = {{http://dx.doi.org/10.1109/3DV62453.2024.00053}}, doi = {{10.1109/3DV62453.2024.00053}}, year = {{2024}}, }