Research map / 3D vision / Human pose estimation
Minimal Solvers for Relative Pose Estimation of Multi-Camera Systems using Affine Correspondences
Banglei Guan, Ji Yin Zhao, Dániel Baráth, Friedrich Fraundorfer
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use 2AC vertical solver instead of 8pt-Kneip | relative pose RANSAC · KITTI sequence 00 | rotation error (deg): 0.130 -> 0.031 | Improved |
| use 2AC vertical solver instead of 4pt-Sweeney | relative pose RANSAC · nuScenes Part 1 | translation error (deg): 1.659 -> 1.469 | Improved |
| use 2AC plane solver instead of 6pt-Stewenius under forward highway motion | relative pose RANSAC · KITTI sequence 01 | translation error (deg): 41.19 -> 2.486 | Improved |
| use 2AC plane instead of 8pt-Kneip on non-planar real driving data | relative pose RANSAC · KITTI sequence 00 | rotation error (deg): 0.130 -> 0.280 | Hurt |
| use two ACs instead of one AC under planar motion | 1AC plane solver · planar motion noise | translation error | Improved |
| add a second AC to avoid equal camera altitude degeneracy | 1AC plane solver · KITTI stereo rig | pose recoverability | Improved |
| increase affine support square side from 20 to 40 pixels (ablation) | AC-based solvers · synthetic data | pose error: 20 -> 40 | Improved |
| use true ACs instead of hallucinated PCs derived from ACs (ablation) | PC-based solvers · synthetic random motion with vertical direction | pose error | Improved |
| shrink hallucinated PC distribution area from 10 to 1 pixels (ablation) | PC-based solvers · synthetic data | pose error: w = 10 -> w = 1 | Improved |
| use 2AC vertical solver instead of 17pt-Li in RANSAC | relative pose RANSAC · KITTI runtime | runtime (s): 52.82 -> 0.09 | Improved |
| use 2AC plane instead of 6pt-Stewenius | single solver · solver runtime | runtime: 3275.4 -> 3.6 microseconds | Improved |
| add known vertical direction prior with IMU noise around 0.4 degrees | 2AC vertical · synthetic random motion | pose error: 0.4 | Improved |
Closest papers by meaning
- Minimal Cases for Computing the Generalized Relative Pose Using Affine Correspondences (ICCV 2021 · similarity 0.98)
- Affine Correspondences Between Multi-Camera Systems for Relative Pose Estimation (TPAMI 2025 · similarity 0.94)
- A Complete Solution to Generalized Relative Pose Estimation From Affine Correspondences (TPAMI 2026 · similarity 0.94)
- From Correspondences to Pose: Non-minimal Certifiably Optimal Relative Pose without Disambiguation (CVPR 2024 · similarity 0.89)
- Practical Solutions to the Relative Pose of Three Calibrated Cameras (CVPR 2025 · similarity 0.88)
- Efficient Solution of Point-Line Absolute Pose (CVPR 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
