A Multi-Source Fusion Lightweight Tightly Coupled LiDAR-Inertial-GNSS SLAM for Unmanned Aerial Vehicle Detection

Autonomous outdoor inspections using unmanned aerial vehicles (UAVs) are often repeated using RTK positioning technology combined with path-planning algorithms. However, positioning is often ineffective when RTK visibility is poor or strong electromagnetic interference exists near the UAV. Light Detection and Ranging (LiDAR)-based simultaneous positioning and mapping (SLAM) technology combines RTK and INS information to form a multi-source fusion system that can provide robust positioning results for UAVs during RTK visibility interruptions. This paper proposes a lightweight, tightly coupled LiDAR-Inertial-GNSS SLAM framework, FLI3D, designed for indoor, outdoor low-altitude and high-altitude inspection scenarios. This framework constructs a Lidar-inertial system based on the state-of-the-art point-plane direct registration method. It introduces a lightweight Bow3D bag-of-words model as a loop closure detection solution to eliminate cumulative error. Additionally, the algorithm provides a GNSS signal fusion interface based on global factor graph optimization, allowing the system to integrate GNSS global observation information such as GPS and RTK signals, and employs a novel balanced fusion strategy to reduce overall computational resources for backend optimization. We conducted extensive experiments on datasets such as NCLT and MCD to validate the proposed method, and the results demonstrate that our system achieves high operational speed and robust performance.