arXiv subjects

Yiling Yao

Publications and source records attributed to Yiling Yao.

3 recordsLinked to original sources

arXiv2026

STARS-GS: Structure-Aware Regularized Gaussian Splatting for Large-Scale Aerial Surface Reconstruction

Large-scale 3D surface reconstruction from aerial imagery is fundamental to geospatial mapping and urban modeling. Recent advances in 3D Gaussian Splatting (3DGS) have demonstrated considerable potential for this task. However, existing methods still face three major challenges in large and complex scenes: scene partitioning may split continuous scene elements across independently optimized sub-regions; geometric constraints mainly focus on the attributes of individual Gaussians while overlooking their local organization; and uniform regularization struggles to accommodate heterogeneous geometric structures. To address these issues, we propose STARS-GS, a structure-aware 3DGS framework for large-scale surface reconstruction. First, we introduce a structure-aware scene partitioning strategy that better preserves continuous scene structures during partitioning and reduces cross-region geometric inconsistencies and stitching artifacts through boundary refinement. Second, we develop neighborhood-aware Gaussian organization that extends geometric constraints from individual primitives to their neighborhood organization, encouraging Gaussians to better conform to local surface geometry. Third, we introduce adaptive surface regularization that adjusts the regularization strength according to local geometric characteristics, promoting geometric consistency in structured regions while preserving plausible variations in unstructured regions. Extensive experiments on large-scale aerial photogrammetry benchmarks demonstrate that STARS-GS consistently outperforms the evaluated Gaussian-based methods in surface reconstruction. It increases the average F1-score from 0.640 for the second-best method to 0.698, corresponding to a relative improvement of approximately 9.1\%, demonstrating effective improvements in geometric accuracy and surface completeness.

cs.CV↗

arXiv2026

BRF-GS: Hyperspectral Bidirectional Reflectance Factor Modeling and Image Generation Based on 3D Gaussian Splatting

The bidirectional reflectance factor (BRF) characterizes the directional radiative properties of terrestrial surfaces. However, existing three-dimensional (3D) radiative transfer models require complex scene construction and computationally intensive radiative transfer solvers, limiting efficient generation of multi-angle hyperspectral reflectance imagery. 3D Gaussian Splatting (3DGS) offers an efficient framework for neural scene representation and novel view synthesis, but its low-order spherical harmonics representation is insufficient for complex directional reflectance, while the high dimensionality and inter-band quality differences of hyperspectral data introduce additional challenges. To address these challenges, we propose BRF-GS, a 3DGS-based framework for BRF modeling and hyperspectral reflectance image generation. BRF-GS introduces a hybrid BRDF-driven kernel to represent complex directional reflectance, selects geometry-reliable spectral bands for robust 3D scene initialization, and adopts a two-stage training strategy that decouples geometry optimization from spectral modeling. We further construct the AIR-BRF dataset, a multi-angle hyperspectral directional reflectance dataset comprising three scenes with diverse natural and artificial targets. Experiments demonstrate that BRF-GS achieves superior spatial and spectral fidelity and accurately reproduces characteristic view-dependent BRF responses. The proposed framework provides an efficient data-driven approach for BRF modeling and multi-angle hyperspectral reflectance image generation in remote sensing scenes.

cs.CV↗

arXiv2024

ARSGaussian: 3D Gaussian Splatting with LiDAR for Aerial Remote Sensing Novel View Synthesis

Novel View Synthesis (NVS) can reconstruct scenes from multi-view images and synthesize novel images from new viewpoints, which provides technical support for tasks such as target recognition and environmental perception. Aerial remote sensing can conveniently capture a wealth of multi-view images with just a few flights. However, the challenges brought by large distances and sparse viewing angles during collection can cause the model to easily produce floaters and overgrowth issues due to geometric estimation errors. This results in low visual quality and a lack of precise geometric estimation capabilities. Therefore, this study presents ARSGaussian, an innovative novel view synthesis (NVS) method for aerial remote sensing. The method incorporates LiDAR point cloud as constraints into the 3D Gaussian Splatting approach, adaptively guiding the Gaussians to grow and split along geometric benchmarks, thereby addressing the overgrowth and floaters issues. Additionally, considering the geometric distortions arising from data acquisition, coordinate transformations with distortion parameters are integrated to replace the simple pinhole camera model parameters to achieve pixel-level alignment between LiDAR point cloud and multi-view optical images, facilitating the accurate fusion of heterogeneous data and achieving the high-precision geo-alignment. Moreover, depth, normal and scale consistency losses are introduced into the regularization process to guide Gaussians toward real depth and plane representations, significantly improving geometric estimation accuracy. To address the current lack of dense airborne hybrid datasets, we have established and released AIR-LONGYAN, an open-source dataset containing a dense LiDAR point cloud (8 pts/m) and multi-view optical images captured by airborne scanners and cameras in diverse scenes....

cs.CV↗