Robust indoor localization via factor graph fusion with motion-aware regression, adaptive relocalization, and continuous map constraints

Kuang, Yujin, Liu, Jiang, Wang, Zhengdong, Meng, Xiangyin, Yang, Yuan, Zhang, Xiaoguo and Liu, Hantao ORCID: https://orcid.org/0000-0003-4544-3481 2026. Robust indoor localization via factor graph fusion with motion-aware regression, adaptive relocalization, and continuous map constraints. Engineering Applications of Artificial Intelligence 179 , 115195. 10.1016/j.engappai.2026.115195

Full text not available from this repository.

Abstract

Accurate and robust indoor localization remains a major challenge due to sensor noise, motion ambiguity, and the absence of reliable absolute references. This study proposes a robust multi-modal localization framework that fuses inertial, visual, and geometric map cues within a unified factor graph optimization. A Motion-Aware Velocity Regression Network (MAVRNet) is first introduced to regress pedestrian velocity directly from raw inertial measurements while adaptively encoding motion contexts. To address visual relocalization instability, a fuzzy-logic-based adaptive random sample consensus is developed to dynamically select inliers according to image entropy and feature density. Furthermore, environmental architectural priors are formulated as a continuous signed distance field and integrated as optimizable soft-penalty constraints in the factor graph, enabling long-term drift suppression and smooth structural consistency. Extensive evaluations in complex indoor scenarios demonstrate that the proposed system achieves a median positioning error of 0.43 meters, significantly outperforming state-of-the-art visual–inertial baselines. Notably, the framework exhibits robust resilience against severe sensory degradation, highlighting its readiness for scalable deployment in large-scale navigation and digital twin infrastructures.

Item Type:	Article
Date Type:	Publication
Status:	Published
Schools:	Schools > Computer Science & Informatics
Publisher:	Elsevier BV
ISSN:	0952-1976
Date of Acceptance:	20 May 2026
Last Modified:	04 Aug 2026 22:04
URI:	https://orca.cardiff.ac.uk/id/eprint/187461

Actions (repository staff only)

Edit Item