CSS-BA: Gate-Guided Column Space Search for Bundle Adjustment

2026-07-17

AI Summary

This work addresses the instability in pose and calibration estimation commonly observed in traditional Schur-complement-based Levenberg-Marquardt bundle adjustment (BA) under weak-geometry scenarios—such as low parallax or near-pure rotation—where ill-conditioning degrades performance. The authors propose a plug-and-play solver-level enhancement that preserves the original BA objective function and trust-region framework while introducing a geometry-aware gating mechanism to restrict parameter updates to a low-dimensional column space. This approach innovatively integrates column space search (CSS) with geometric gating, constraining only the update direction without eliminating any variables. Experimental results demonstrate that the method significantly improves optimization stability, relative pose accuracy, and calibration performance across both general and weak-geometry settings, all while maintaining excellent reprojection quality.

Abstract

Bundle adjustment (BA) remains a critical refinement module for image-based 3D reconstruction and continues to improve geometric accuracy even in learning-based pipelines. However, in low-parallax and near-rotational regimes, classical Schur-based Levenberg--Marquardt (LM) often becomes ill-conditioned and yields unreliable pose and calibration estimates. We propose Gate-Guided CSS-BA, a solver-side modification of Schur-LM that preserves the classical BA objective and trust-region framework while constraining each update to a geometrically informed low-dimensional subspace. By integrating Column Space Search (CSS) with geometry-aware gating, the method stabilizes the Schur-LM update without altering the estimation problem. In contrast to keyframe or state-selection approaches, all camera and point parameters remain in the optimization problem; only the update direction is restricted. The method serves as a drop-in replacement for existing BA pipelines. Experiments on both generic and challenging weak-geometry scenarios show more stable optimization, improved relative pose accuracy, and competitive calibration behavior while maintaining reprojection quality.

Problem

• bundle adjustment
• ill-conditioned
• low-parallax
• near-rotational
• pose estimation

Innovation

• Bundle Adjustment
• Column Space Search
• Schur complement
• Geometric gating
• Levenberg-Marquardt