TY  - JOUR
AU  - HU, Jihua
PY  - 2026
TI  - Gaussian splatting modeling design integrating oblique and close photogrammetry
JO  - Experimental Technology and Management
SN  - 1002-4956
SP  - 214
EP  - 219
VL  - 43
IS  - 10
AB  - ObjectiveIn the context of constructing three-dimensional (3D) Digital China, photogrammetry from unmanned aerial vehicles (UAVs) has become a key technology for real-scene 3D modeling and an important practical part of education for surveying and mapping. Traditional oblique photogrammetry is limited by high-altitude viewing angles and occlusions created by buildings, which leads to obvious model holes, geometric distortions, blurred textures, and missing details in occluded areas such as building facades and under eaves. Although close-range photogrammetry can supplement high-precision data for these blind areas, the combined modeling of oblique and close-range images suffers from serious bundle adjustment conflicts, pose drift, and matching failures due to huge differences in scale, viewpoint, and resolution. Traditional mesh-based models have difficulties in local editing, region replacement, and seamless fusion, which result in high complexity, poor reproducibility, and unsuitability for standardized teaching of experimental techniques. To solve these problems, in this paper we propose an integrated 3D modeling experimental scheme based on 3D Gaussian Splatting (3DGS) for combining oblique and close-range photogrammetry.MethodsIn this research, we have designed a modular modeling and coordinate-based replacement scheme. First, we obtain five-directional oblique photogrammetry carried out at a flight height of 110 m to acquire large-scale images that cover the building roof, main structure, and ground, and we construct a global 3D Gaussian model from these data. Second, we perform close-range photogrammetry using parallel facade-following routes at a distance of about 5 m to capture high-resolution images of facades, under-eaves spaces, and windows; this enables us to generate a refined local Gaussian model with complete details. Taking advantage of the discrete, editable, and seamlessly fusible features of 3DGS, we reconstruct the global and local models separately to avoid the bundle adjustment conflicts caused by heterogeneous images. We next establish a unified geographic coordinate transformation framework to align the two local coordinate systems. By computing the origin offsets between the main model and the facade model in geographic space, we are able to crop and remove the defective facade areas in the main model and map the refined Gaussian particles accurately onto the corresponding positions of the main model. We implement this replacement purely through calculations of translations, thus ensuring high spatial alignment and natural fusion without requiring manual repairs, mesh modifications, or texture remapping.ResultsWe carried out experiments on a campus teaching building using an RTK UAV and DJI Terra software that supports 3D Gaussian Splatting. We collected a total of 421 oblique images and 84 close-range images. The original main model generated by oblique photogrammetry performs well on the roof and ground but has obvious defects on the facade. In contrast, the close-range modeling result presents complete and clear details of the facade without holes or distortions. After the coordinate transformation and Gaussian particle replacement, the final integrated model achieves a complete structure, with sharp edges, full under-eaves details, clear windows, and no geometric abnormalities or texture artifacts. The whole process is stable, efficient, and highly reproducible.ConclusionsThis method fundamentally solves the problem of modeling defects that is inherent in traditional oblique photogrammetry of occluded regions and avoids the fusion conflicts in combined close-range photogrammetry. It simplifies the modeling workflow, reduces the necessity for manual repairs, and forms a standardized experimental procedure that is suitable for practical teaching. The integrated experiment covers UAV flight planning, data acquisition, 3D reconstruction, coordinate transformation, and Gaussian model editing, which can effectively cultivate students’ practical abilities in UAV mapping and real-scene 3D modeling. This method provides a new technical scheme and teaching mode for high-precision 3D modeling, and it can be widely used in teaching the experimental methods involved in surveying and mapping, obtaining geographic information, remote sensing, and UAV application technology.
UR  - https://doi.org/10.16791/j.cnki.sjg.2026.10.028
DO  - 10.16791/j.cnki.sjg.2026.10.028