Autonomous Drone Navigation for Structural Inspection Using 3D Gaussian Splatting and Optimization-based Path Planning

Structural inspections increasingly rely on unmanned aerial vehicles (UAVs) to improve access and safety, however, UAV-based bridge inspections often suffer from inconsistent data coverage, pilot-dependent flight paths, and limited repeatability across inspection campaigns. This study presents a fully autonomous drone inspection framework that integrates 3D Gaussian Splatting (3DGS) with optimization-based path planning using a 3D A* algorithm. The framework reconstructs a photorealistic 3D model from video data to identify Regions of Interest and automatically generates collision-free flight trajectories that balance surface coverage, flight distance, and inspection time. Field test on the Low Level Bridge in Edmonton, Canada demonstrates that a high-fidelity 3DGS preview can be produced in around five minutes, enabling rapid on-site rendering. The optimized helical path achieves 91.4% coverage with a 45.8% reduction in mission time, offering a scalable, model-driven solution for safe and efficient structural health monitoring.