​​Integrating Geometric Priors and Importance Sampling for High-fidelity Indoor Scene Reconstruction

doi:10.16182/j.issn1004731x.joss.25-1055

Abstract

Abstract:

Gaussian splatting suffers from geometric distortion during scene reconstruction, particularly in weakly textured indoor scenes. To address this issue, this paper proposes a high-precision indoor scene reconstruction method that integrates geometric priors and importance sampling. The proposed method fully considers the effect of the initialization process on reconstruction quality. An advanced feed-forward model is employed to generate high-quality geometric initialization, thus improving overall reconstruction stability and accuracy. An importance sampling strategy is introduced to mitigate the adverse effects of blurry images. Furthermore, a supervision mechanism based on a geometric prior model is designed to constrain the scene structure, further enhancing geometric consistency and reconstruction accuracy. Experimental results show that the proposed method improves reconstruction quality and effectively alleviates geometric structural distortion in indoor scene reconstruction.

Key words: 3D reconstruction, Gaussian splatting, indoor scene, feed-forward model, geometric prior, surface reconstruction

CLC Number:

TP37

Yang Tao, Shi Min, Zhao Xigang, Wang Suqin, Wang Qi, Zhu Dengming. Integrating Geometric Priors and Importance Sampling for High-fidelity Indoor Scene Reconstruction[J]. Journal of System Simulation, 2026, 38(3): 584-594.

Figures/Tables 7

Fig. 1

Fig. 2

Fig. 3

Table 1

Table 2

Fig. 4

Table 3

References 28

[1]	王浩宇, 龚光红, 蔡继红, 等. 基于战场元宇宙的动态三维场景感知[J]. 系统仿真学报, 2023, 35(10): 2262-2278.
	Wang Haoyu, Gong Guanghong, Cai Jihong, et al. Dynamic 3D Scene Perception Based on Battlefield Metaverse[J]. Journal of System Simulation, 2023, 35(10): 2262-2278.
[2]	王晓宇. 基于虚拟现实技术的分布式三维室内设计系统设计[J]. 现代电子技术, 2019, 42(12): 183-186.
	Wang Xiaoyu. Design of Distributed 3D Indoor Design System Based on Virtual Reality Technology[J]. Modern Electronics Technique, 2019, 42(12): 183-186.
[3]	郑国威, 贾鹏, 李晓飞, 等. ZED双目相机点云数据融合与重建下的室内导航SLAM优化算法[J]. 系统仿真技术, 2025, 21(1): 62-67.
	Zheng Guowei, Jia Peng, Li Xiaofei, et al. Optimization Algorithm for Indoor Navigation SLAM Based on Point Cloud Data Fusion and Reconstruction of ZED Binocular Camera[J]. System Simulation Technology, 2025, 21(1): 62-67.
[4]	代超. 建筑工程测绘中的三维激光扫描技术应用分析[J]. 城市建设理论研究(电子版), 2025(9): 179-181.
[5]	Snavely N, Seitz S M, Szeliski R. Photo Tourism: Exploring Photo Collections in 3D[J]. ACM Transactions on Graphics, 2006, 25(3): 835-846.
[6]	Galliani Silvano, Lasinger Katrin, Schindler Konrad. Massively Parallel Multiview Stereopsis by Surface Normal Diffusion[C]//Proceedings of the 2015 IEEE International Conference on Computer Vision (ICCV). USA: IEEE Computer Society, 2015: 873-881.
[7]	Schönberger Johannes L, Zheng Enliang, Frahm J M, et al. Pixelwise View Selection for Unstructured Multi-View Stereo[C]//Computer Vision-ECCV 2016. Cham: Springer International Publishing, 2016: 501-518.
[8]	Tola Engin, Strecha Christoph, Fua Pascal. Efficient Large-scale Multi-view Stereo for Ultra High-resolution Image sets[J]. Machine Vision and Applications, 2012, 23(5): 903-920.
[9]	Mildenhall B, Srinivasan P P, Tancik M, et al. NeRF: Representing Scenes as Neural Radiance Fields for View Synthesis[J]. Communications of the ACM, 2022, 65(1): 99-106.
[10]	Osher S, Sethian J A. Fronts Propagating with Curvature-dependent Speed: Algorithms Based on Hamilton-Jacobi Formulations[J]. Journal of Computational Physics, 1988, 79(1): 12-49.
[11]	Kerbl Bernhard, Kopanas Georgios, Leimkuehler Thomas, et al. 3D Gaussian Splatting for Real-time Radiance Field Rendering[J]. ACM Transactions on Graphics, 2023, 42(4): 139.
[12]	Yu Zehao, Chen Anpei, Huang Binbin, et al. Mip-splatting: Alias-free 3D Gaussian Splatting[C]//2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2024: 19447-19456.
[13]	Yang Haosen, Zhang Chenhao, Wang Wenqing, et al. Improving Gaussian Splatting with Localized Points Management[C]//2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2025: 21696-21705.
[14]	Lu Tao, Yu Mulin, Xu Linning, et al. Scaffold-GS: Structured 3D Gaussians for View-adaptive Rendering[C]//2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2024: 20654-20664.
[15]	Yariv L, Gu J, Kasten Y, et al. Volume Rendering of Neural Implicit Surfaces[J]. Advances in Neural Information Processing Systems, 2021, 34: 4805-4815.
[16]	Wang Peng, Liu Lingjie, Liu Yuan, et al. NeuS: Learning Neural Implicit Surfaces by Volume Rendering for Multi-view Reconstruction[C]//Proceedings of the 35th International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2021: 27171-27183.
[17]	Li Zhaoshuo, Müller Thomas, Evans A, et al. Neuralangelo: High-fidelity Neural Surface Reconstruction[C]//2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2023: 8456-8465.
[18]	Yu Zehao, Peng Songyou, Niemeyer M, et al. MonoSDF: Exploring Monocular Geometric Cues for Neural Implicit Surface Reconstruction[C]//Proceedings of the 36th International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2022: 25018-25032.
[19]	Guédon Antoine, Lepetit Vincent. SuGaR: Surface-aligned Gaussian Splatting for Efficient 3D Mesh Reconstruction and High-quality Mesh Rendering[C]//2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2024: 5354-5363.
[20]	Zhang Baowen, Fang Chuan, Shrestha R, et al. RaDe-GS: Rasterizing Depth in Gaussian Splatting[EB/OL]. (2024-06-24) [2025-09-05]. .
[21]	Huang Binbin, Yu Zehao, Chen Anpei, et al. 2D Gaussian Splatting for Geometrically Accurate Radiance Fields[C]//ACM SIGGRAPH 2024 Conference Papers. New York: Association for Computing Machinery, 2024: 32.
[22]	Leroy Vincent, Cabon Yohann, Revaud Jerome. Grounding Image Matching in 3D with MASt3R[C]//Computer Vision-ECCV 2024. Cham: Springer Nature Switzerland, 2025: 71-91.
[23]	Bochkovskii A, Delaunoy Amaël, Germain H, et al. Depth Pro: Sharp Monocular Metric Depth in Less Than a Second[EB/OL]. (2025-04-21) [2025-08-26]. .
[24]	Benesty Jacob, Chen Jingdong, Huang Yiteng, et al. Pearson Correlation Coefficient[M]//Israel Cohen, Huang Yiteng, Chen Jingdong, et al. Noise Reduction in Speech Processing. Berlin: Springer Berlin Heidelberg, 2009: 1-4.
[25]	Ye Chongjie, Qiu Lingteng, Gu Xiaodong, et al. StableNormal: Reducing Diffusion Variance for Stable and Sharp Normal[J]. ACM Transactions on Graphics, 2024, 43(6): 250.
[26]	Dai A, Chang A X, Savva M, et al. ScanNet: Richly-annotated 3D Reconstructions of Indoor Scenes[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2017: 2432-2443.
[27]	Straub J, Whelan T, Ma Lingni, et al. The Replica Dataset: A Digital Replica of Indoor Spaces[EB/OL]. (2019-06-13) [2025-08-16]. .
[28]	Wang Jiepeng, Wang Peng, Long Xiaoxiao, et al. NeuRIS: Neural Reconstruction of Indoor Scenes Using Normal Priors[C]//Computer Vision-ECCV 2022. Cham: Springer Nature Switzerland, 2022: 139-155.

方法	准确率	完整率	精确率	召回率	F1分数	训练时间/min
SuGaR	0.081 2	0.101 8	0.591 4	0.507 0	0.515 6	120.67
RaDe-GS	0.063 8	0.076 1	0.618 2	0.631 1	0.623 8	13.65
2DGS	0.076 1	0.108 6	0.675 1	0.577 1	0.622 2	17.96
本文算法	0.021 7	0.050 6	0.936 8	0.793 7	0.859 2	30.43

方法	准确率	完整率	精确率	召回率	F1分数	训练时间/min
SuGaR	0.115 4	0.115 9	0.419 0	0.400 6	0.409 0	86.51
RaDe-GS	0.125 7	0.152 0	0.380 7	0.410 1	0.392 6	7.43
2DGS	0.149 8	0.123 4	0.427 1	0.433 4	0.428 9	8.86
本文算法	0.063 8	0.046 8	0.733 8	0.693 3	0.711 4	15.12

方法	准确率	完整率	精确率	召回率	F1分数
基线模型	0.076 1/0.149 8	0.108 6/0.123 4	0.675 1/0.427 1	0.577 1/0.433 4	0.622 2/0.428 9
+前馈模型初始化	0.024 7/0.081 1	0.062 3/0.063 3	0.896 6/0.634 3	0.749 2/0.627 9	0.816 2/0.629 4
+重要图片采样	0.024 3/0.076 8	0.061 6/0.058 0	0.918 1/0.677 9	0.771 8/0.659 4	0.838 5/0.666 8
+几何先验监督(完整模型)	0.021 7/0.063 8	0.050 6/0.046 8	0.938 6/0.733 8	0.793 7/0.693 3	0.859 2/0.711 4