Enhanced VINS-Fusion-Based SLAM Algorithm for Reliable Camera Pose Estimation in GNSS-Challenging Environments

Abstract

Simultaneous Localization and Mapping (SLAM) is a prominent research area in robotics. This paper presents a SLAM algorithm based on VINS-Fusion to address the challenge of estimating camera pose in GNSS-challenging environments. The algorithm incorporates a method for evaluating the reliability of GNSS measurements to optimize their usage, and performs online optimization of the GNSS lever arm. These enhancements improve the robustness and positioning accuracy of the system in GNSS-challenging scenarios. The GNSS measurement quality evaluation method consists of two layers: the first layer is situated at the input end of GNSS measurements and utilizes a dynamic threshold method to monitor the quality of the input measurements; the second layer is employed after the system completes graph optimization and employs a chi-square test to identify all GNSS outliers present in the optimization graph. Experimental results demonstrate that the proposed algorithm outperforms the VINS-Fusion algorithm significantly in GNSS-challenging environments. In dataset testing, the root mean square error (RMSE) is reduced by an average of 76.3%.