Geometric-texture tightly-coupled factor graph SLAM algorithm

This paper proposes a tightly-coupled multi-modal sensor fusion framework, GT-TC-FGS, which achieves high-precision real-time state estimation through deep fusion of LiDAR, monocular camera, and IMU. The framework employs an Error State Iterative Kalman Filter (ESIKF) as its core architecture, combined with factor graph optimization to construct a hierarchical fusion system. To address the challenge of loop closure detection in complex scenarios, this work innovatively introduces a Geometric-Texture Joint Consistency Verification (GT-JCV) method. By integrating LiDAR geometric features and visual texture features, GT-JCV establishes a dual-validation mechanism through multi-modal spatial structure constraints and appearance similarity matching, significantly enhancing the reliability of loop closure detection. Extensive experiments on public KITTI and M2DGR datasets demonstrate that GT-TC-FGS outperforms state-of-the-art LiDAR-visual-inertial SLAM algorithms (e.g., LVI-SAM) in terms of accuracy and robustness.