Abstract
Accurate and robust localization is critical for autonomous unmanned systems, particularly in challenging environments where conventional Global Navigation Satellite System (GNSS) integrated navigation can be highly susceptible to degeneration. This paper presents a trajectory cross-verification method for loosely coupled Visual-Inertial Navigation System (VINS)-GNSS integration system in GNSS degeneration environments. Our method does not depend on GNSS covariance or single-point measurements. Instead, it performs VINS and GNSS trajectory alignment and further cross-verification through a dual sliding window filtering algorithm. Our method decouples horizontal and vertical data, minimizing the adverse effects of invalid vertical data on the alignment and verification results. Furthermore, we propose a gravity-aware vertical verification strategy to ensure robust vertical trajectory alignment and outlier filtering. We evaluated our method both public and our real-world datasets, and compared it against both classic and state-of-the-art (SOTA) approaches. The experimental results demonstrate that our trajectory cross-verification algorithm exhibits accurate and robust localization performance in GNSS degeneration scenarios, outperforming SOTA algorithms in most datasets.
Funding Information
  • Chinese National Programs for Fundamental Strengthening Research (2020-jcjq-zd-133-11)
  • Key Laboratory of Smart Earth (KF2023YB01-10)