High-Fidelity 3D Reconstruction for Open-Pit Mine Digital Twins Using UAV Data and an Integrated 3D Gaussian Splatting Pipeline

Addressing the challenges in 3D reconstruction of large-scale open-pit mines, such as dramatic terrain undulations, complex texture features, and the difficulty of balancing geometric accuracy with real-time rendering efficiency using traditional methods, this paper proposes a high-fidelity reconstruction framework integrating UAV multi-modal data with the state-of-the-art 3D Gaussian Splatting (3DGS) architecture. First, an integrated air-ground multi-modal data acquisition system is established. Using a UAV equipped with LiDAR and a high-resolution camera, high-quality geometric and textural data of the mining area are acquired through terrain-adaptive flight planning. Second, to tackle the VRAM bottlenecks and loose geometric structures inherent in original 3DGS for large scenes, we adopt the advanced CityGaussianV2 architecture as our core reconstruction engine. By leveraging its divide-and-conquer parallel training strategy, 2DGS planar geometric constraints, and Decomposed Gradient Densification (DGD) mechanism, this framework effectively overcomes memory limitations and significantly enhances the geometric sharpness of slope crests and toes. Finally, engineering validation was conducted at Kambove Mining. Experimental results demonstrate that the proposed method achieves centimeter-level geometric accuracy, a real-time web rendering frame rate exceeding 60 FPS, and a model storage compression rate of over 90%. The digital twin control platform built upon this model successfully achieves deep fusion and visual scheduling of multi-source heterogeneous data, providing a novel technical path for constructing high-precision reality-based foundations for smart mines.