Incremental Structure-from-Motion through Affine Transformations of Monocular Depth Priors
(2025) In Master's Theses in Mathematical Sciences FMAM05 20251Mathematics (Faculty of Engineering)
- Abstract
- This thesis presents an incremental Structure-from-Motion pipeline through the use of affine transformations of monocular depth priors. The scene points are triangulated by averaging lifted covisible keypoints utilizing the calculated transformations. Each new image is registered to the reconstruction by first estimating the camera pose, then aligning the depth prior to the observed 3D points. We propose a bundle adjustment method based on reprojection errors and depth deviations through uniform sampling across multi-view shared 2D-2D correspondences. The final reconstruction is found through backprojection of the registered images using the transformed depth priors. We achieve improved completeness and F1 score compared to the widely... (More)
- This thesis presents an incremental Structure-from-Motion pipeline through the use of affine transformations of monocular depth priors. The scene points are triangulated by averaging lifted covisible keypoints utilizing the calculated transformations. Each new image is registered to the reconstruction by first estimating the camera pose, then aligning the depth prior to the observed 3D points. We propose a bundle adjustment method based on reprojection errors and depth deviations through uniform sampling across multi-view shared 2D-2D correspondences. The final reconstruction is found through backprojection of the registered images using the transformed depth priors. We achieve improved completeness and F1 score compared to the widely adopted COLMAP pipeline, but show that this comes at the cost of accuracy. Furthermore, we observe improved robustness and accuracy in terms of pose estimation for low three-view overlap in images. (Less)
Please use this url to cite or link to this publication:
https://lup.lub.lu.se/student-papers/record/9194789
- author
- Olsteg, Isac LU
- supervisor
-
- Viktor Larsson LU
- Gustav Hanning LU
- organization
- course
- FMAM05 20251
- year
- 2025
- type
- H2 - Master's Degree (Two Years)
- subject
- publication/series
- Master's Theses in Mathematical Sciences
- report number
- 2025:E50
- ISSN
- 1404-6342
- other publication id
- LUTFMA-3592-2025
- language
- English
- id
- 9194789
- date added to LUP
- 2025-09-15 11:11:45
- date last changed
- 2026-02-12 09:25:27
@misc{9194789,
abstract = {{This thesis presents an incremental Structure-from-Motion pipeline through the use of affine transformations of monocular depth priors. The scene points are triangulated by averaging lifted covisible keypoints utilizing the calculated transformations. Each new image is registered to the reconstruction by first estimating the camera pose, then aligning the depth prior to the observed 3D points. We propose a bundle adjustment method based on reprojection errors and depth deviations through uniform sampling across multi-view shared 2D-2D correspondences. The final reconstruction is found through backprojection of the registered images using the transformed depth priors. We achieve improved completeness and F1 score compared to the widely adopted COLMAP pipeline, but show that this comes at the cost of accuracy. Furthermore, we observe improved robustness and accuracy in terms of pose estimation for low three-view overlap in images.}},
author = {{Olsteg, Isac}},
issn = {{1404-6342}},
language = {{eng}},
note = {{Student Paper}},
series = {{Master's Theses in Mathematical Sciences}},
title = {{Incremental Structure-from-Motion through Affine Transformations of Monocular Depth Priors}},
year = {{2025}},
}