Publications
ワークショップ (国際) Turntable-Constrained Camera Pose Estimation
Norio Kosaka, Shinichi Higashino, Shuji Yamaguchi
The IEEE/CVF Conference on Computer Vision and Pattern Recognition 2026 (CVPR 2026)
2026.6.7
We study structure-from-motion (SfM) for image sequences generated by single-axis rotation (turntable or orbiting-camera motion), where per-view rotation angles are unknown (potentially non-uniform) and camera intrinsics may be known, partially known, or unknown. While generic SfM pipelines estimate unconstrained pairwise geometry and only later attempt to fit an orbit, the single-axis assumption defines a low-dimensional, physically interpretable motion family in which all views share a common rotation axis and differ primarily by a single scalar angle. We argue that exploiting this structure \emph{during} estimation is essential for physically consistent reconstructions, and we illustrate a concrete failure mode of an off-the-shelf pipeline even on perfectly single-axis data. This workshop draft formalises the problem setting, situates it relative to classic turntable geometry and recent orbit-constrained optimisation, and outlines a multi-stage estimation strategy (constrained initialisation, turntable-constrained epipolar fitting, and turntable bundle adjustment), leaving the final minimal solver and robust objective open. Finally, we provide a small synthetic evaluation protocol and baseline suite for quantifying axis and angle recovery.
Paper :
Turntable-Constrained Camera Pose Estimation
（外部サイト）