2026· IEEE Transactions on Instrumentation and Measurement· Vol 75, pp. 5020218-5020218· 0 citations· 64 references
TL;DR
A robust SCR-based visual localization framework that improves camera pose measurement by modeling feature-level uncertainty and geometric consistency is proposed and a co-visibility feature interpolation strategy is proposed to simulate continuous camera viewpoint perturbations and improve robustness against appearance variations.
Abstract
Visual localization is a key technology in vision-based instrumentation and measurement systems, where the six-degree-of-freedom (6-DoF) pose of a calibrated camera is estimated with respect to a scene-fixed reference coordinate system. Among various approaches, scene coordinate regression (SCR) provides an efficient solution by directly regressing the 3-D scene coordinates corresponding to 2-D pixels and subsequently using perspective-n-point (PnP) solvers to determine the camera pose. Despite promising performance in small-scale scenarios, SCR methods can face localization accuracy challenges in large-scale or visually complex environments. Such challenges can arise from ambiguous local features and global representations that are sensitive to appearance variations. In this article, we propose a robust SCR-based visual localization framework that improves camera pose measurement by modeling feature-level uncertainty and geometric consistency. First, we introduce an uncertainty-guided feature sampling mechanism to explicitly evaluate the reliability of local features and prioritize high-quality observations during training. Second, we optimize a lightweight linear projection layer via a geometry-adaptive multisimilarity (GA-MS) loss to enforce geometric consistency among co-visible images in the feature space. Furthermore, we propose a co-visibility feature interpolation (CFI) strategy to simulate continuous camera viewpoint perturbations and improve robustness against appearance variations. Experimental results demonstrate that the proposed method achieves improved accuracy and reliability compared to state-of-the-art visual localization approaches on both indoor and outdoor benchmark datasets. Additional measurement-oriented analyses and real-world validation further support its reliability and applicability to practical visual pose measurement.
Visual relocalization requires compact scene representations that capture global spatial context while preserving local geometric precision. Existing regression-based methods often emphasize one aspect at the expense of the other. Global image representations encode scene-level spatial information but provide limited...
Ying Ye, Rui Wei, Chang Jiang et al.· Frontiers of Mechanical Engi...· 0 citations
Localizing an event camera against a pre-built LiDAR map can be cast as dense optical-flow estimation between a rendered depth view and an event image, followed by a Perspective-n-Point (PnP) solver over the induced 3D-2D correspondences. Existing pipelines rely on geometric consensus during pose estimation, but do not...
Panagiotis Kiousis, Kuang-Yi Chen, Jun Zhang et al.· 0 citations
Visual odometry (VO) estimates camera motion from image sequences and is essential for robotics, autonomous driving, and AR/VR. Robust VO remains challenging because large viewpoint changes and strong parallax make reliable cross-frame motion cues difficult to capture, especially in the presence of visual disturbances...
Jun-Qi Bao, Qing-Ying Wu, Jun Huang et al.· IEEE Transactions on Instrum...· 0 citations
Omnidirectional cameras provide surround-view information, significantly overcoming the field-of-view constraints for traditional visual localization systems. However, existing localization systems fail to fully exploit the characteristics of omnidirectional cameras, struggling to deliver accurate and robust localizati...
Zhuo Yu, Shu-Yang Zhang, Hui-Hua Zhu et al.· 0 citations
A Temporal-aware Localization framework (TempLoc) designed to enhance the robustness of outdoor localization by effectively modeling sequential consistency is proposed, demonstrating the effectiveness of temporal-aware correspondence modeling in LiDAR relocalization.
Ming-Hang Zhu, Zhi-Jing Wang, Yu-Xin Guo et al.· Proceedings of the 34th ACM...· 0 citations