Jul 2026· The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences· 0 citations· 18 references
Abstract
Abstract. The proliferation of continuous Neural Radiance Field (NeRF) and 3D Gaussian Splatting (3DGS) has shifted the paradigm of 3D aerial reconstruction from relying solely on geometric stereo matching to inverse rendering optimization. However, while these emerging rendering-based frameworks excel in synthesizing photo-realistic novel views, their capability to extract accurate surfaces in complex aerial scenarios remains ambiguous compared to traditional methods. To establish a clearer understanding, this study presents a comprehensive evaluation of five representative frameworks spanning traditional Structure from Motion (SfM), purely Signed Distance Field (SDF) representations, unstructured 3D Gaussians, hybrid voxel-Gaussians, and strictly explicit sparse voxels. By systematically standardizing identical computational environments, inputs, and unified mesh-extraction pipelines on both real-world airborne LiDAR datasets and synthetic cityscapes, we assess their performance regarding visual fidelity, geometric accuracy, and resource efficiency. The experimental results reveal that while traditional MVS produces the highest overall geometric precision by strictly enforcing multi-view rigid geometry, it is prone to failures in texture-less regions. Among rendering-based representations, a fundamental trade-off exists: highly flexible, unstructured 3DGS achieve highest visual scores but degrade the underlying geometric surfaces; conversely, explicitly structured techniques demonstrate distinct superiority in regularizing topological coherence and floating artifact suppression. Furthermore, we observe that integrating structured voxels avoids the severe memory bottlenecks associated with extracting geometries from chaotic unorganized primitives. These empirical findings emphasize that for large-scale aerial photogrammetry, integrating explicit spatial structuralization into differentiable rendering pipelines is imperative for achieving scalable operations and bridging the geometric accuracy gap with traditional methods.
Abstract. Accurate 3D reconstruction of Cultural Heritage (CH) assets remains a challenging task when scenes include complex geometries and non-Lambertian surfaces, such as dense vegetation, reflective ceramics, or polished materials, which often degrade the performance of traditional multi-view stereo (MVS) pipelines. This work investigates the potential of Mesh-In-the-Loop Gaussian Splatting (MILo), a recent extension of 3D Gaussian Splatting (3DGS) that integrates differentiable mesh extraction directly within the optimization process, enabling bidirectional consistency between volumetric and surface representations. The method is evaluated on three challenging CH datasets: a monumental Tilia tomentosa tree located in a UNESCO-listed urban garden, a reflective ceramic object from the Sarreguemines Earthenware Museum and the South Portal façade of the Notre-Dame Cathedral of Strasbourg. MILo-based reconstructions are compared against standard photogrammetric MVS results generated with Agisoft Metashape, using terrestrial laser scanner (TLS) point clouds as geometric reference. Quantitative accuracy assessment is performed through Multiscale Model-to-Model Cloud Comparison (M3C2), focusing on error distribution, standard deviation, outlier percentage, and preservation of fine-scale structures. Results indicate that while conventional MVS performs slightly better on stable architectural surfaces, MILo significantly improves reconstruction consistency for complex organic geometries, substantially reducing outliers and better preserving thin structures. These findings highlight the suitability of MILo for CH documentation scenarios characterized by challenging surface properties and intricate natural forms.
D. Billi, Chaimaa Delasse, Arnadi Murtiyoso et al.· The International Archives o...· 0 citations
3D Gaussian Splatting is effective in multi-view surface reconstruction tasks, fundamental to photogrammetry applications. Mainstream pipelines focus on optimizing for explicit Gaussians to enhance volumetric rendering quality, typically relying on rendering-based Truncated Signed Distance Function (TSDF) methods for reconstruction, while the training and meshing often suffer from missing details and geometric distortions under occlusions or restricted viewpoints. In this paper, we focus on a new problem definition: directly formulating implicit surface representation exploiting geometric features of trained Gaussian Surfels. For this problem, we introduce Gaussian Surfels with Spatial Awareness (GSSA) for reconstruction, a spatially aware reconstruction framework that optimizes for Surfel distributions from geometric primitives via a 3D refiner, and constructs SDF by integrating signed distances from voxel vertices to nearby Surfels. Compared with state-of-the-art GS-based, especially Surfel-based, reconstruction methods, GSSA achieves both competitive geometric accuracy and efficiency. On terrestrial (indoor object) and airborne (photogrammetry) reconstruction benchmarks, GSSA maintains a competitive balance between reconstruction time and accuracy. Further migration experiments demonstrate that GSSA can also be integrated as a plug-and-play module into general Surfel-based pipelines.
Hao Tang, Siyuan Zou, Hongbo Pan et al.· Remote Sensing· 0 citations
Abstract. High-resolution 3D documentation of cultural heritage sites is essential for their preservation. While terrestrial laser scanning (TLS) remains the gold standard, it is often cost-intensive compared to photogrammetry. This study evaluates three image-based reconstruction techniques, Multi-View Stereo (MVS), Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS), by applying them to a complex scene featuring a chapel and its surrounding vegetation, sensed from an uncrewed aerial vehicle (UAV). A hybrid TLS/MVS model provides a high-accuracy reference. Using identical interior and exterior camera parameters of the 105 UAV-acquired images, we generate dense point clouds with all methods and assess geometric accuracy and completeness using the M3C2 algorithm. Results show that MVS achieves superior accuracy (standard deviation of all M3C2 distances: MVS = 0.11 m, NeRF = 0.15 m), whereas NeRF attains up to 20% higher completeness, particularly in low-texture and vegetation-occluded regions. The 3DGS point cloud was deemed too sparse and was therefore not used for further analysis. The study highlights the potential of NeRFs to recover partially occluded or sparsely textured geometries that are challenging for MVS and suggests a complementary use of both approaches for cost-efficient documentation of cultural heritage.
Frederik Schulte, P. Akwensi, L. Winiwarter· The International Archives o...· 0 citations
A semantic-guided 3D Gaussian splatting (3DGS) framework tailored to sparse-view industrial reconstruction was introduced, enabling robust reconstruction from limited viewpoints and offers a practical geometric foundation for automated inspection and remote equipment monitoring.
Boyang Li, Tianhan Gao, Zuan Gu et al.· Visual Computing for Industr...· 0 citations
High-precision 3D reconstruction of objects with complex surfaces, such as ancient architecture and detailed artworks, requires close-range image acquisition, which remains challenging for Unmanned Aerial Vehicle (UAV) systems. The operational proximity of current UAV workflows is often insufficient to capture fine geometric and textural details, limiting high-fidelity digitization. This paper presents a georeferenced NeRF-based UAV acquisition framework for automated waypoint planning and supervised close-proximity execution. The core of the framework is a path-planning module that operates on a metric geometric prior established through Geographic Neural Radiance Fields (Geo-NeRF), which denotes a georeferenced NeRF modeling pipeline rather than a new NeRF architecture or loss function. By generating waypoints directly on this neural representation and optimizing the flight path via a nearest-neighbor strategy, the proposed framework supports close-proximity image acquisition for static targets under controlled conditions. Empirical validation demonstrates improved close-range flight proximity, photographic accuracy, and 3D reconstruction fidelity compared with the evaluated baselines.
Haoyu Liu, Yizhi Zou, Lu Yang et al.· Drones· 0 citations
Abstract. To address the issues of geometric distortion and loss of details in 3D modeling for complex cultural heritage scenes, this paper proposes an improved 3D Gaussian Splatting (3DGS) reconstruction method that integrates LiDAR and illumination-awareness. First, high-precision 3D coordinates from LiDAR point clouds are utilized to guide the initialization of Gaussian Primitives, establishing a precise geometric foundation and effectively overcoming deformation on weakly textured surfaces. Second, an illumination-aware network is constructed to dynamically adjust appearance parameters by combining global illumination from images with LiDAR reflectance intensity. This decouples complex lighting from material properties, accurately reproducing the unique textures of artifacts. Finally, a multi-dimensional joint loss function incorporating photometric, scale, and appearance smoothness constraints is introduced to collaboratively optimize scene geometry, appearance, and camera poses. Experimental results on indoor and outdoor cultural heritage preservation scenarios demonstrate that the proposed method significantly outperforms various comparative algorithms in terms of both visual fidelity and geometric accuracy. The quantitative and qualitative evaluations confirm that our approach effectively eliminates geometric distortions and recovers fine texture details, providing an efficient and reliable technical solution for the digital preservation of cultural heritage.
Xiao Liu, Xinying Li, Wan Li et al.· ISPRS Annals of the Photogra...· 0 citations
