基于高斯泼溅的复杂场景重建研究现状与展望

doi:10.16182/j.issn1004731x.joss.25-1144

系统仿真学报 ›› 2026, Vol. 38 ›› Issue (6): 1535-1566.doi: 10.16182/j.issn1004731x.joss.25-1144

基于高斯泼溅的复杂场景重建研究现状与展望

李洪安¹^,², 杨嘉乐¹, 刘庆芳³, 石郁⁴

^1.西安科技大学人工智能与计算机学院，陕西西安 710054
^2.北京航空航天大学虚拟现实技术与系统全国重点实验室，北京 100191
^3.中国邮政集团有限公司培训中心，河北石家庄 050021
^4.西安市公安局未央分局，陕西西安 710016

收稿日期:2025-11-20 修回日期:2026-02-10 出版日期:2026-06-25 发布日期:2026-06-26
通讯作者: 刘庆芳
第一作者简介:李洪安(1978-)，男，副教授，博士，研究方向为计算机图形学、可视媒体计算。
基金资助:
虚拟现实技术与系统全国重点实验室(北京航空航天大学)开放课题(VRLAB2023B08);陕西省自然科学基础研究计划(2025JC-YBMS-679);山西省水利科学技术研究推广项目(2024GM13);邮政应用技术协同创新中心资助项目(YB2025004)

Current Status and Prospects of Complex Scene Reconstruction Based on Gaussian Splatting

Li Hong'an¹^,², Yang Jiale¹, Liu Qingfang³, Shi Yu⁴

^1.College of Artificial Intelligence & Computer Science, Xi'an University of Science and Technology, Xi'an 710054, China
^2.State Key Laboratory of Virtual Reality Technology and Systems, Beihang University, Beijing 100191, China
^3.Training Center of China Post Group, Shijiazhuang 050021, China
^4.Weiyang Sub-Bureau of Xi'an Public Security Bureau, Xi'an 710016, China

Received:2025-11-20 Revised:2026-02-10 Online:2026-06-25 Published:2026-06-26
Contact: Liu Qingfang

摘要/Abstract

摘要：

三维高斯泼溅（3D Gaussian splatting， 3DGS）从显式表示的角度为新视角合成提供了另一种思路，使用3D高斯基元重建场景并通过基于点的光栅化过程替代传统的光线积分，不仅提高了训练和渲染效率，也为复杂场景重建提供了新的思路。将基于3DGS的复杂场景重建方法划分三大类，围绕大规模场景、稀疏视角和动态场景这3类方向进行展开描述，回顾了该领域的发展现状，并指出未来可能的研究方向。

关键词: 三维高斯泼溅, 复杂场景, 大规模场景, 稀疏视角, 动态场景

Abstract:

Three-dimensional Gaussian splatting (3DGS) provides an alternative approach for novel view synthesis from the perspective of explicit representation. By reconstructing scenes using 3D Gaussian primitives and replacing traditional ray integration with a point-based rasterization process, it not only improves training and rendering efficiency but also offers new insights for complex scene reconstruction. This paper divided 3DGS-based complex scene reconstruction methods into three major categories and elaborated on them around large-scale scenes, sparse views, and dynamic scenes. It reviewed the current development status of this field and pointed out possible future research directions.

Key words: 3D Gaussian splatting, complex scene, large-scale scene, sparse view, dynamic scene

中图分类号:

TP391.4

李洪安,杨嘉乐,刘庆芳等 . 基于高斯泼溅的复杂场景重建研究现状与展望[J]. 系统仿真学报, 2026, 38(6): 1535-1566.

Li Hong'an,Yang Jiale,Liu Qingfang,et al . Current Status and Prospects of Complex Scene Reconstruction Based on Gaussian Splatting[J]. Journal of System Simulation, 2026, 38(6): 1535-1566.

图/表 43

图1

表1

表2

表 3

表 4

表5

常用评价指标

指标	取值范围	评价内容
PSNR	$[0, + ∞)$	衡量像素精度，数值越高表示整体重建质量越好
SSIM	$[- 1, + 1]$	衡量结构一致性，数值越高表示纹理与结构还原度更高
LPIPS	$[0,1]$	衡量语义一致性，数值越低表示语义特征更接近真实图像

表5

图2

图3

图4

图5

图6

图7

图8

表6

表7

表8

表9

表10

图9

图10

图11

图12

图13

图14

图15

图16

图17

图18

表11

表12

表13

图19

图20

图21

图22

图23

图 24

图25

图26

表14

表15

图27

表16

参考文献 103

[1]	Li Mingrui, Liu Shuhong, Deng Tianchen, et al. DenseSplat: Densifying Gaussian Splatting SLAM with Neural Radiance Prior[J]. IEEE Transactions on Visualization and Computer Graphics, 2026, 32(2): 1993-2006.
[2]	Li Mingrui, He Jiaming, Wang Yangyang, et al. End-to-end RGB-D SLAM with Multi-MLPs Dense Neural Implicit Representations[J]. IEEE Robotics and Automation Letters, 2023, 8(11): 7138-7145.
[3]	Liu Shuhong, Deng Tianchen, Zhou Heng, et al. MG-SLAM: Structure Gaussian Splatting SLAM with Manhattan World Hypothesis[J]. IEEE Transactions on Automation Science and Engineering, 2025, 22: 17034-17049.
[4]	Li Mingrui, Guo Zhetao, Deng Tianchen, et al. DDN-SLAM: Real Time Dense Dynamic Neural Implicit SLAM[J]. IEEE Robotics and Automation Letters, 2025, 10(5): 4300-4307.
[5]	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.
[6]	Garbin S J, Kowalski M, Johnson M, et al. FastNeRF: High-fidelity Neural Rendering at 200FPS[C]//2021 IEEE/CVF International Conference on Computer Vision (ICCV). Piscataway: IEEE, 2021: 14326-14335.
[7]	Müller Thomas, Evans A, Schied C, et al. Instant Neural Graphics Primitives with a Multiresolution Hash Encoding[J]. ACM Transactions on Graphics, 2022, 41(4): 102.
[8]	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.
[9]	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.
[10]	朱东林, 陈淼, 毛宇岩, 等. 三维高斯泼溅技术在场景重建中的研究现状与挑战[J]. 集成技术, 2025, 14(4): 1-20.
	Zhu Donglin, Chen Miao, Mao Yuyan, et al. Research Status and Challenges of 3D Gaussian Splatting Technology in Scene Reconstruction[J]. Journal of Integration Technology, 2025, 14(4): 1-20.
[11]	高建, 陈林卓, 沈秋, 等. 基于三维高斯溅射技术的可微分渲染研究进展(特邀) [J]. 激光与光电子学进展, 2024, 61(16): 153-165.
	Gao Jian, Chen Linzhuo, Shen Qiu, et al. Advances in Differentiable Rendering Based on Three-dimensional Gaussian Splatting(Invited) [J]. Laser & Optoelectronics Progress, 2024, 61(16): 153-165.
[12]	Fei Ben, Xu Jingyi, Zhang Rui, et al. 3D Gaussian Splatting as a New Era: A Survey[J]. IEEE Transactions on Visualization and Computer Graphics, 2025, 31(8): 4429-4449.
[13]	Bao Yanqi, Ding Tianyu, Huo Jing, et al. 3D Gaussian Splatting: Survey, Technologies, Challenges, and Opportunities[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2025, 35(7): 6832-6852.
[14]	Schönberger Johannes L, Frahm J M. Structure-from-motion Revisited[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2016: 4104-4113.
[15]	Fisher A, Cannizzaro R, Cochrane M, et al. ColMap: A Memory-efficient Occupancy Grid Mapping Framework[J]. Robotics and Autonomous Systems, 2021, 142: 103755.
[16]	Turki H, Ramanan D, Satyanarayanan M. Mega-NeRF: Scalable Construction of Large-scale NeRFs for Virtual Fly-throughs[C]//2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2022: 12912-12921.
[17]	Lin Liqiang, Liu Yilin, Hu Yue, et al. Capturing, Reconstructing, and Simulating: The UrbanScene3D Dataset[C]//Computer Vision-ECCV 2022. Cham: Springer Nature Switzerland, 2022: 93-109.
[18]	Li Yixuan, Jiang Lihan, Xu Linning, et al. MatrixCity: A Large-scale City Dataset for City-scale Neural Rendering and Beyond[C]//2023 IEEE/CVF International Conference on Computer Vision (ICCV). Piscataway: IEEE, 2023: 3182-3192.
[19]	Mildenhall B, Srinivasan P P, Ortiz-Cayon R, et al. Local Light Field Fusion: Practical View Synthesis with Prescriptive Sampling Guidelines[J]. ACM Transactions on Graphics, 2019, 38(4): 29.
[20]	Barron J T, Mildenhall B, Verbin D, et al. Mip-NeRF 360: Unbounded Anti-aliased Neural Radiance Fields[C]//2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2022: 5460-5469.
[21]	Jensen Rasmus, Dahl Anders, Vogiatzis G, et al. Large Scale Multi-view Stereopsis Evaluation[C]//2014 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2014: 406-413.
[22]	Pumarola Albert, Corona Enric, Pons-Moll Gerard, et al. D-NeRF: Neural Radiance Fields for Dynamic Scenes[C]//2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2021: 10313-10322.
[23]	Park K, Sinha U, Barron J T, et al. Nerfies: Deformable Neural Radiance Fields[C]//2021 IEEE/CVF International Conference on Computer Vision (ICCV). Piscataway: IEEE, 2021: 5845-5854.
[24]	Park K, Sinha U, Hedman P, et al. HyperNeRF: A Higher-dimensional Representation for Topologically Varying Neural Radiance Fields[EB/OL]. (2021-09-10) [2025-11-20]. .
[25]	Wang Liao, Zhang Jiakai, Liu Xinhang, et al. Fourier PlenOctrees for Dynamic Radiance Field Rendering in Real-time[C]//2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2022: 13514-13524.
[26]	Li Tianye, Slavcheva M, Zollhoefer M, et al. Neural 3D Video Synthesis from Multi-view Video[C]//2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2022: 5511-5521.
[27]	Shin Yoon J, Kim K, Gallo O, et al. Novel View Synthesis of Dynamic Scenes with Globally Coherent Depths from a Monocular Camera[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2020: 5335-5344.
[28]	Broxton M, Flynn J, Overbeck R, et al. Immersive Light Field Video with a Layered Mesh Representation[J]. ACM Transactions on Graphics, 2020, 39(4): 86.
[29]	Sabater N, Boisson G, Vandame B, et al. Dataset and Pipeline for Multi-view Light-field Video[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). Piscataway: IEEE, 2017: 1743-1753.
[30]	Wang Zhou, Bovik A C, Sheikh H R, et al. Image Quality Assessment: From Error Visibility to Structural Similarity[J]. IEEE Transactions on Image Processing, 2004, 13(4): 600-612.
[31]	Zhang R, Isola P, Efros A A, et al. The Unreasonable Effectiveness of Deep Features as a Perceptual Metric[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2018: 586-595.
[32]	Lin Jiaqi, Li Zhihao, Tang Xiao, et al. VastGaussian: Vast 3D Gaussians for Large Scene Reconstruction[C]//2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2024: 5166-5175.
[33]	Liu Yang, Luo Chuanchen, Fan Lue, et al. CityGaussian: Real-time High-quality Large-scale Scene Rendering with Gaussians[C]//Computer Vision-ECCV 2024. Cham: Springer Nature Switzerland, 2025: 265-282.
[34]	Chen Yu, Lee G H. DOGS: Distributed-oriented Gaussian Splatting for Large-scale 3D Reconstruction via Gaussian Consensus[C]//Proceedings of the 38th International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2024: 34487-34512.
[35]	Wu Yongchang, Qi Zipeng, Shi Zhenwei, et al. BlockGaussian: Efficient Large-scale Scene Novel View Synthesis via Adaptive Block-based Gaussian Splatting[EB/OL]. (2025-04-15) [2025-11-20]. .
[36]	Fan Jixuan, Li Wanhua, Han Yifei, et al. Momentum-GS: Momentum Gaussian Self-distillation for High-quality Large Scene Reconstruction[C]//2025 IEEE/CVF International Conference on Computer Vision (ICCV). Piscataway: IEEE, 2025: 25250-25260.
[37]	Feng Guofeng, Chen Siyan, Fu Rong, et al. FlashGS: Efficient 3D Gaussian Splatting for Large-scale and High-resolution Rendering[C]//2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2025: 26652-26662.
[38]	Gao Yuanyuan, Li Hao, Chen Jiaqi, et al. CityGS-X: A Scalable Architecture for Efficient and Geometrically Accurate Large-scale Scene Reconstruction[EB/OL]. (2025-03-29) [2025-11-20]. .
[39]	Li Bingling, Chen Shengyi, Wang Luchao, et al. RetinaGS: Scalable Training for Dense Scene Rendering with Billion-scale 3D Gaussians[EB/OL]. (2024-06-22) [2025-11-20]. .
[40]	Chen Junyi, Ye Weicai, Wang Yifan, et al. GigaGS: Scaling up Planar-based 3D Gaussians for Large Scene Surface Reconstruction[EB/OL]. (2024-09-10) [2025-11-20]. .
[41]	Song Kaiwen, Zeng Xiaoyi, Ren Chenqu, et al. City-on-web: Real-time Neural Rendering of Large-scale Scenes on the Web[C]//Computer Vision-ECCV 2024. Cham: Springer Nature Switzerland, 2025: 385-402.
[42]	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.
[43]	Ren Kerui, Jiang Lihan, Lu Tao, et al. Octree-GS: Towards Consistent Real-time Rendering with LOD-structured 3D Gaussians[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence. (2025-05-08) [2025-11-20]. .
[44]	Kerbl Bernhard, Meuleman Andreas, Kopanas Georgios, et al. A Hierarchical 3D Gaussian Representation for Real-time Rendering of Very Large Datasets[J]. ACM Transactions on Graphics, 2024, 43(4): 62.
[45]	Yuan Zhensheng, Huang Haozhi, Xiong Zhen, et al. Robust and Efficient 3D Gaussian Splatting for Urban Scene Reconstruction[C]//2025 IEEE/CVF International Conference on Computer Vision (ICCV). Piscataway: IEEE, 2025: 26209-26219.
[46]	Kulhanek Jonas, Peng Songyou, Kukelova Zuzana, et al. WildGaussians: 3D Gaussian Splatting in the Wild[EB/OL]. (2024-10-31) [2025-11-20]. .
[47]	Zhang Dongbin, Wang Chuming, Wang Weitao, et al. Gaussian in the Wild: 3D Gaussian Splatting for Unconstrained Image Collections[C]//Computer Vision- ECCV 2024. Cham: Springer Nature Switzerland, 2025: 341-359.
[48]	Zhang Chenhao, Cao Yuanping, Zhang Lei. CrossView-GS: Cross-view Gaussian Splatting For Large-scale Scene Reconstruction[EB/OL]. (2025-04-28) [2025-11-20]. .
[49]	Li Zhuoxiao, Yao Shanliang, Wu Taoyu, et al. ULSR-GS: Urban Large-scale Surface Reconstruction Gaussian Splatting with Multi-view Geometric Consistency[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2025, 230: 861-880.
[50]	Zhang Saining, Ye Baijun, Chen Xiaoxue, et al. Drone-assisted Road Gaussian Splatting with Cross-view Uncertainty[EB/OL]. (2024-08-27) [2025-11-20]. .
[51]	Jiang Lihan, Ren Kerui, Yu Mulin, et al. Horizon-Gs: Unified 3D Gaussian Splatting for Large-scale Aerial-to-ground Scenes[C]//2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2025: 26789-26799.
[52]	Huang Binbin, Yu Zehao, Chen Anpei, et al. 2D Gaussian Splatting for Geometrically Accurate Radiance Fields[C]//ACM SIGGRAPH 2024 conference papers. New York: ACM, 2024: 32.
[53]	Liu Y, Luo C, Mao Z, et al. CityGaussianV2: Efficient and Geometrically Accurate Reconstruction for Large-scale Scenes[C]//Proceedings of the Thirteenth International Conference on Learning Representations(ICLR)[S.l.: s.n.], 2025:88324-84341.
[54]	Xu Haofei, Peng Songyou, Wang Fangjinhua, et al. DepthSplat: Connecting Gaussian Splatting and Depth[C]//2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2025: 16453-16463.
[55]	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.
[56]	Gao Yuanyuan, Dai Yalun, Li Hao, et al. CoSurfGS: Collaborative 3D Surface Gaussian Splatting with Distributed Learning for Large Scene Reconstruction[EB/OL]. (2024-12-23) [2025-11-20]. .
[57]	Mi Zhenxing, Xu D. Switch-nerf: Learning Scene Decomposition with Mixture of Experts for Large-scale Neural Radiance Fields[C]//The Eleventh International Conference on Learning Representations (ICLR). OpenReview [S.l.: s.n.], 2022.
[58]	Zhao Chen, Wang Xuan, Zhang Tong, et al. Self-ensembling Gaussian Splatting for Few-shot Novel View Synthesis[C]//2025 IEEE/CVF International Conference on Computer Vision (ICCV). Piscataway: IEEE, 2025: 4940-4950.
[59]	Chung Jaeyoung, Oh Jeongtaek, Mu Lee Kyoung. Depth-regularized Optimization for 3D Gaussian Splatting in Few-shot Images[C]//2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). Piscataway: IEEE, 2024: 811-820.
[60]	Li Jiahe, Zhang Jiawei, Bai Xiao, et al. DNGaussian: Optimizing Sparse-view 3D Gaussian Radiance Fields with Global-local Depth Normalization[C]//2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2024: 20775-20785.
[61]	Zhu Zehao, Fan Zhiwen, Jiang Yifan, et al. FSGS: Real-time Few-shot View Synthesis Using Gaussian Splatting[C]//Computer Vision-ECCV 2024. Cham: Springer Nature Switzerland, 2025: 145-163.
[62]	Zheng Y, Jiang Z, He S, et al. NexusGS: Sparse View Synthesis with Epipolar Depth Priors in 3D Gaussian Splatting[C]//2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2025: 26800-26809.
[63]	黄志勇, 佘雅丽, 华喜锋, 等. DCSplat: 一种深度约束的稀疏视角三维重建方法[J]. 图学学报, 2025, 46(3): 510-519.
	Huang Zhiyong, She Yali, Hua Xifeng, et al. DCSplat: Gaussian Splatting with Depth Information Constraints Under Sparse Viewpoints[J]. Journal of Graphics, 2025, 46(3): 510-519.
[64]	Zhang Jiawei, Li Jiahe, Yu Xiaohan, et al. CoR-GS: Sparse-view 3D Gaussian Splatting via Co-regularization[C]//Computer Vision-ECCV 2024. Cham: Springer Nature Switzerland, 2025: 335-352.
[65]	Han Liang, Zhou Junsheng, Liu Yushen, et al. Binocular-guided 3D Gaussian Splatting with View Consistency for Sparse View Synthesis[C]//Proceedings of the 38th International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2024: 68595-68621.
[66]	Xu Yexing, Wang Longguang, Chen Minglin, et al. DropoutGS: Dropping Out Gaussians for Better Sparse-view Rendering[C]//2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2025: 701-710.
[67]	Park Hyunwoo, Ryu Gun, Kim Wonjun. DropGaussian: Structural Regularization for Sparse-view Gaussian Splatting[C]//2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2025: 21600-21609.
[68]	Song Meixi, Lin Xin, Zhang Dizhe, et al. D² GS: Depth-and-density Guided Gaussian Splatting for Stable and Accurate Sparse-view Reconstruction[EB/OL]. (2025-10-09) [2025-11-20]. .
[69]	Wan Yecong, Shao Mingwen, Cheng Yuanshuo, et al. S2Gaussian: Sparse-view Super-resolution 3D Gaussian Splatting[C]//2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2025: 711-721.
[70]	Hu Xinyuan, Shi Changyue, Yang Chuxiao, et al. SRSplat: Feed-forward Super-resolution Gaussian Splatting from Sparse Multi-view Images[EB/OL]. (2025-11-15) [2025-11-20]. .
[71]	Dai Dongrui, Xing Yuxiang. EAP-GS: Efficient Augmentation of Pointcloud for 3D Gaussian Splatting in Few-shot Scene Reconstruction[C]//2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2025: 16498-16507.
[72]	Sun Xiangyu, Chen Runnan, Gong Mingming, et al. Intern-GS: Vision Model Guided Sparse-view 3D Gaussian Splatting[EB/OL]. (2025-05-27) [2025-11-20]. .
[73]	Yu Hanyang, Long Xiaoxiao, Tan Ping. LM-Gaussian: Boost Sparse-view 3D Gaussian Splatting with Large Model Priors[EB/OL]. (2025-07-28) [2025-11-20]. .
[74]	Lu Xiaohan, Fu Jiaye, Zhang Jiaqi, et al. ProSplat: Improved Feed-forward 3D Gaussian Splatting for Wide-baseline Sparse Views[EB/OL]. (2025-06-09) [2025-11-20]. .
[75]	Zhou Junsheng, Zhang Weiqi, Liu Yushen. DiffGS: Functional Gaussian Splatting Diffusion[C]//Proceedings of the 38th International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2024: 37535-37560.
[76]	Chen Yuedong, Zheng Chuanxia, Xu Haofei, et al. MVSplat360: Feed-forward 360 Scene Synthesis from Sparse Views[C]//Proceedings of the 38th International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2024: 107064-107086.
[77]	Chen Kangjie, Dai Bingquan, Qin Minghan, et al. SLGaussian: Fast Language Gaussian Splatting in Sparse Views[C]//Proceedings of the 33rd ACM International Conference on Multimedia. New York: ACM, 2025: 3047-3056.
[78]	He Zongqi, Xiao Zhe, Chan K C, et al. See in Detail: Enhancing Sparse-view 3D Gaussian Splatting with Local Depth and Semantic Regularization[C]//2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). Piscataway: IEEE, 2025: 1-5.
[79]	Liao Guibiao, Li Qing, Bao Zhenyu, et al. SPC-GS: Gaussian Splatting with Semantic-prompt Consistency for Indoor Open-world Free-view Synthesis from Sparse Inputs[C]//2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2025: 11264-11274.
[80]	Niemeyer Michael, Barron J T, Mildenhall B, et al. RegNeRF: Regularizing Neural Radiance Fields for View Synthesis from Sparse Inputs[C]//2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2022: 5470-5480.
[81]	Yang Jiawei, Pavone M, Wang Yue. FreeNeRF: Improving Few-shot Neural Rendering with Free Frequency Regularization[C]//2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2023: 8254-8263.
[82]	Yang Zeyu, Pan Zijie, Zhu Xiatian, et al. 4D Gaussian Splatting: Modeling Dynamic Scenes with Native 4D Primitives[EB/OL]. (2025-08-05) [2025-11-20]. .
[83]	Yang Ziyi, Gao Xinyu, Zhou Wen, et al. Deformable 3D Gaussians for High-fidelity Monocular Dynamic Scene Reconstruction[C]//2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2024: 20331-20341.
[84]	Fridovich-Keil S, Meanti Giacomo, Frederik Rahbæk Warburg, et al. K-planes: Explicit Radiance Fields in Space, Time, and Appearance[C]//2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2023: 12479-12488.
[85]	Cao Ang, Johnson J. HexPlane: A Fast Representation for Dynamic Scenes[C]//2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2023: 130-141.
[86]	Wu Guanjun, Yi Taoran, Fang Jiemin, et al. 4D Gaussian Splatting for Real-time Dynamic Scene Rendering[C]//2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2024: 20310-20320.
[87]	Lin Youtian, Dai Zuozhuo, Zhu Siyu, et al. Gaussian-flow: 4D Reconstruction with Dynamic 3D Gaussian Particle[C]//2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2024: 21136-21145.
[88]	Liang Yiqing, Khan N, Li Zhengqin, et al. GauFRe: Gaussian Deformation Fields for Real-time Dynamic Novel View Synthesis[C]//2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV). Piscataway: IEEE, 2025: 2642-2652.
[89]	Zhao Guosheng, Wang Xiaofeng, Ni Chaojun, et al. ReconDreamer++: Harmonizing Generative and Reconstructive Models for Driving Scene Representation[EB/OL]. (2025-07-10) [2025-11-20]. .
[90]	Sun Su, Zhao Cheng, Sun Zhuoyang, et al. SplatFlow: Self-supervised Dynamic Gaussian Splatting in Neural Motion Flow Field for Autonomous Driving[C]//2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2025: 27487-27496.
[91]	Wu Jiahao, Peng Rui, Wang Zhiyan, et al. Swift 4D: Adaptive Divide-and-conquer Gaussian Splatting for Compact and Efficient Reconstruction of Dynamic Scene[EB/OL]. (2025-03-16) [2025-11-20]. .
[92]	Li Ziwen, Huang Jiaxin, Chen Runnan, et al. UrbanGS: Semantic-guided Gaussian Splatting for Urban Scene Reconstruction[EB/OL]. (2025-03-21) [2025-11-20]. .
[93]	Zhang Xinjie, Liu Zhening, Zhang Yifan, et al. MEGA: Memory-efficient 4D Gaussian Splatting for Dynamic Scenes[C]//2025 IEEE/CVF International Conference on Computer Vision (ICCV). Piscataway: IEEE, 2025: 27828-27838.
[94]	Yuan Y, Shen Q, Yang X, et al. 1000+FPS 4D Gaussian Splatting for Dynamic Scene Rendering[C]//The Thirty-ninth Annual Conference on Neural Information Processing Systems (NeurIPS)[S.l.: s.n.], 2025.
[95]	Oh Cho Woong, Cho In, Kim Seoha, et al. 4D Scaffold Gaussian Splatting with Dynamic-aware Anchor Growing for Efficient and High-fidelity Dynamic Scene Reconstruction[EB/OL]. (2025-08-05) [2025-11-20]. .
[96]	Chan Lee Joo, Jong Hwan Ko, Park Eunbyung. Optimized Minimal 3D Gaussian Splatting[EB/OL]. (2025-11-06) [2025-11-20]. .
[97]	Song Rui, Liang Chenwei, Xia Yan, et al. CoDa-4DGS: Dynamic Gaussian Splatting with Context and Deformation Awareness for Autonomous Driving[C]//2025 IEEE/CVF International Conference on Computer Vision (ICCV). Piscataway: IEEE, 2025: 28031-28041.
[98]	Huang Yiming, Cui Beilei, Bai Long, et al. Endo-4DGS: Endoscopic Monocular Scene Reconstruction with 4D Gaussian Splatting[C]//Medical Image Computing and Computer Assisted Intervention-MICCAI 2024. Cham: Springer Nature Switzerland, 2024: 197-207.
[99]	Jiang Hanxiao, Hsu H Y, Zhang Kaifeng, et al. PhysTwin: Physics-informed Reconstruction and Simulation of Deformable Objects from Videos[EB/OL]. (2025-03-23) [2025-11-20]. .
[100]	Cai Junhao, Yang Yuji, Yuan Weihao, et al. GIC: Gaussian-informed Continuum for Physical Property Identification and Simulation[C]//Proceedings of the 38th International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2024: 75035-75063.
[101]	Hong Haoqin, Fan Ding, Dou Fubin, et al. Physics-informed Deformable Gaussian Splatting: Towards Unified Constitutive Laws for Time-evolving Material Field[EB/OL]. (2025-11-09) [2025-11-20]. .
[102]	Xu Jiawei, Fan Zexin, Yang Jian, et al. Grid4D: 4D Decomposed Hash Encoding for High-fidelity Dynamic Gaussian Splatting[C]//Proceedings of the 38th International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2024: 123787-123811.
[103]	彭嘉盈, 莫霄睿, 年成, 等. 基于三维高斯溅射原理的神经渲染硬件加速器研究综述[J]. 微电子学与计算机, 2025, 42(10): 66-77.
	Peng Jiaying, Mo Xiaorui, Cheng Nian, et al. Review on Hardware Accelerator for 3D Gaussian Splatting[J]. Microelectronics & Computer, 2025, 42(10): 66-77.

数据集	图片总数量	覆盖面积/km²	拍摄类型	数据合成	图片分辨率/像素
Mill 19	3.6×10³	125	航拍	真实世界	4 608×3 456
UrbanScene3D	128×10³	136	航拍	真实世界	5 472×3 648
MatrixCity	519×10³	28	航拍	虚拟合成	1 920×1 080

数据集	场景数量	训练视角
LLFF	8	3、6、9
Mip-NeRF360	7	12、24
DTU	15	3、6、9

数据集	场景数量	运动特性	采集设备	纹理/光照	挑战性
D-NeRF	8	物体缓慢运动，复杂几何	合成数据	纯白背景	非刚性变形重建
Nerfies	9	非刚性变形	双同步相机	自然光照	拓扑变化
HyperNeRF	17	刚体/非刚体，大幅运动	双同步相机	自然光照	大幅度运动
NeRF-DS	8	镜面反射，物体运动	双同步相机	高光材质	镜面反射重建

数据集	场景数量	运动特性	采集设备	纹理/光照	挑战性
Plenoptic Video	21	拓扑变化，高光/半透明物体	21台移动GoPro阵列	多光照条件	拓扑变化
NVIDIA	8	人体/物体互动，刚体运动	12台GoPro静态阵列	自然光照	多目标运动
Google Immersive	11	动态街景	46鱼眼相机阵列	自然光照	大规模场景
Light-Field Video	7	面部、中景动画	4×4相机阵列	剧场光照	窄基线视角合成

方法	Building	Rubble	Residence	Sci-Art
CityGaussian	3.07	2.22	2.49	0.88
Momentum-GS	2.45	1.50	2.00	0.97

基于高斯泼溅的复杂场景重建研究现状与展望

Current Status and Prospects of Complex Scene Reconstruction Based on Gaussian Splatting

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 43

参考文献 103

相关文章 5

编辑推荐

Metrics

本文评价

方法	Building	Rubble	Residence	Sci-Art
CityGaussian	8 977	5 527	6 494	2 726
Momentum-GS	5 830	4 106	6 419	6 647

方法	Building		Rubble		Residence		Sci-Art
方法	训练时间	内存/GB	训练时间	内存/GB	训练时间	内存/GB	训练时间	内存/GB
MegaNeRF	19 h 49 min	5.84	30 h 48 min	5.88	27 h 20 min	5.99	27 h 39 min	5.97
SwitchNeRF^[57]	24 h 46 min	5.84	38 h 30 min	5.87	35 h 11 min	5.94	34 h 34 min	5.92
3DGS	21 h 37 min	4.62	18 h 40 min	2.18	23 h 13 min	3.23	21 h 33 min	1.61
VastGaussian	3 h 26 min	3.07	2 h 30 min	2.74	3 h 12 min	3.67	2 h 33 min	3.54
DoGaussian	3 h 51 min	3.39	2 h 25 min	2.54	4 h 33 min	6.11	4 h 23 min	3.53

方法	SSIM	PSNR	LPIPS
D-NeRF	0.975	31.69	0.058
3DGS	0.930	23.19	0.080
HexPlane	0.970	31.04	0.040
K-Planes	0.973	32.32	0.038
4DGS	0.980	34.05	0.020
Gaussian-flow	0.980	34.27	0.030
GauFRe	0.982	34.80	0.020
DeformGS	0.991	39.51	0.012
Grid4D	0.994	42.00	0.008

方法	MS-SSIM	PSNR	t/min
HyperNeRF	0.856	23.47	-
4DGS	0.845	25.20	30
Gaussian-flow	0.862	26.30	12
GauFRe	0.830	24.10	90

大类	子类别	核心设计机理	系统代价	适用边界
大规模场景	空间划分/加速	几何解耦与并行优化	块间一致性维护困难	适用于场景几何基本静态的重建任务，在存在显著动态变化时，块间一致性难以维持
	外观一致性	跨视角/环境光照解耦	增加的外观约束模块	适用于光照条件变化相对缓慢的场景，在剧烈光照变化或频繁曝光切换下效果显著下降
	层级LOD结构	感知驱动的多尺度管理	树结构更新与内存寻址开销	适用于视距分布相对稳定的应用场景，在快速视角切换或频繁尺度跳变时性能受限
稀疏视角	几何正则化	几何先验与一致性约束	重度依赖初始化精度	适用于中低密度观测条件，在极端稀疏视角下重建稳定性明显下降
稀疏视角	生成式/视觉先验	生成式迁移和视角先验知识	推理负荷显著	适用于语义结构相对稳定的场景，对高频细节与真实纹理还原能力有限
动态场景	变形场重建	连续形变重建	底层受限于拓扑连续性	适用于低速、连续形变目标，对拓扑突变或非连续形变适应性不足
	运动重建	显式时空一致性与运动关联	梯度流复杂，计算复杂度大幅上升	适用于中低速运动场景，在高速或强非线性运动条件下易出现不稳定
	高效存储/加速	利用属性量化与冗余剔除	存储开销下降，可能引发精度下降	适用于高实时性需求的应用场景，对重建精度要求较高的任务并不适用
	物理感知重建	引入物理规律优化	参数辨识复杂，计算负荷高	适用于高精度物理一致性重建，难以满足实时或低延迟应用需求

[1]	万飞, 尹勇. 融合GPS先验信息的3D高斯溅射大规模场景配准技术[J]. 系统仿真学报, 2026, 38(3): 563-571.
[2]	李鹏飞, 徐伟, 朴永杰, 方应红, 石敦攀. 天基空间小目标复杂场景数字成像仿真[J]. 系统仿真学报, 2025, 37(9): 2315-2334.
[3]	李维刚, 甘磊, 王永强. 基于LiDAR-IMU的动态场景点云建图方法[J]. 系统仿真学报, 2025, 37(1): 95-106.
[4]	王浩宇, 龚光红, 蔡继红, 叶必鹏, 周照方, 梅铮, 李妮. 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.
[5]	周培培, 侯幸林. 一种用于图像融合的无监督深度神经网络[J]. 系统仿真学报, 2022, 34(6): 1267-1274.

方法	Mill 19 Building			Mill 19 Rubble
方法	SSIM	PSNR	LPIPS	SSIM	PSNR	LPIPS
MegaNeRF	0.547	20.93	0.504	0.553	24.06	0.516
SwitchNeRF	0.579	21.54	0.474	0.562	24.31	0.496
3DGS	0.720	20.46	0.305	0.777	25.47	0.277
VastGaussian	0.804	23.50	0.130	0.823	26.92	0.132
CityGaussian	0.778	21.55	0.246	0.813	25.77	0.228
DoGaussian	0.759	22.73	0.204	0.765	25.78	0.257
Momentum-GS	0.815	23.23	0.194	0.827	25.93	0.201

方法	Residence			Sci-Art
方法	SSIM	PSNR	LPIPS	SSIM	PSNR	LPIPS
MegaNeRF	0.628	22.08	0.489	0.770	25.60	0.390
SwitchNeRF	0.654	22.57	0.457	0.795	26.52	0.360
3DGS	0.791	21.44	0.236	0.830	21.05	0.242
VastGaussian	0.852	24.25	0.124	0.885	26.81	0.121
CityGaussian	0.813	22.00	0.211	0.837	21.39	0.230
DoGaussian	0.740	21.94	0.244	0.804	24.42	0.219
Momentum-GS	0.818	22.21	0.197	0.856	23.02	0.205

方法	3视角			6视角			9视角
方法	SSIM	PSNR	LPIPS	SSIM	PSNR	LPIPS	SSIM	PSNR	LPIPS
RegNeRF^[80]	0.587	19.08	0.336	0.760	23.10	0.206	0.820	24.86	0.161
FreeNeRF^[81]	0.612	19.63	0.308	0.779	23.73	0.195	0.827	25.13	0.160
3DGS	0.649	19.22	0.229	0.814	23.80	0.125	0.860	25.44	0.096
DNGaussian	0.591	19.12	0.294	0.717	22.01	0.246	0.741	22.62	0.244
FSGS	0.682	20.43	0.248	0.823	24.09	0.145	0.860	25.31	0.122
COR-GS	0.712	20.45	0.196	0.837	24.49	0.115	0.874	26.06	0.089
DropGaussian	0.713	20.76	0.200	0.837	24.74	0.117	0.874	26.21	0.088
SE-GS	0.724	20.79	0.183	0.839	24.78	0.110	0.878	26.36	0.084

方法	3视角			6视角			9视角
方法	SSIM	PSNR	LPIPS	SSIM	PSNR	LPIPS	SSIM	PSNR	LPIPS
3DGS	0.804	17.67	0.158	0.894	23.69	0.086	0.941	26.80	0.050
DNGaussian	0.776	18.57	0.178	0.862	22.56	0.114	0.917	25.25	0.077
FSGS	0.822	17.84	0.161	0.905	23.68	0.096	0.941	26.17	0.064
COR-GS	0.835	18.65	0.140	0.910	24.39	0.074	0.950	27.38	0.045
SE-GS	0.857	19.24	0.132	0.924	25.28	0.073	0.958	28.08	0.043