Abstract:
True digital orthophoto maps (DOMs) are vital spatial data sources due to their high precision, detail, and accessibility. However, traditional generation methods using i...Show MoreMetadata
Abstract:
True digital orthophoto maps (DOMs) are vital spatial data sources due to their high precision, detail, and accessibility. However, traditional generation methods using image differential correction with DEM/DSM often produce significant distortions from inaccurate surface data and missing building information. Conventional geometric stitching and radiometric correction methods struggle to improve quality. While neural radiance field (NeRF)-based view synthesis offers progress, issues remain in training efficiency, rendering quality, and scene editability. To address these limitations, we propose Ortho-3DGS, a novel method for orthophoto generation from UAV imagery using 3D Gaussian Splatting (3DGS). Unlike NeRF, our approach models scenes via 3D Gaussian ellipsoids, optimized with depth supervision and gradient-based refinement for explicit and accurate reconstruction. We design a dedicated orthophoto rendering pipeline to generate high-quality, distortion-free DOMs efficiently. Experiments show Ortho-3DGS surpasses traditional tools (ContextCapture, Pix4Dmapper, Metashape) and NeRF-based methods (Ortho-NeRF, Ortho-NGP) in both radiometric and geometric performance. Ortho-3DGS offers commercial-grade accuracy while effectively mitigating distortions and artifacts, especially in complex environments. These results demonstrate its value for fast, precise orthophoto generation in diverse geospatial applications.
Published in: IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing ( Volume: 18)