融合4DGS和AR驱动的地下空间数字模型可视化交付方法

doi:10.11996/JG.j.2095-302X.2026040895

摘要/Abstract

摘要：

随着地下空间工程规模持续扩大与建设复杂度不断提升，传统基于BIM、点云或网格的静态离散模型交付方式，存在更新滞后、数据量大或视觉失真等问题，难以满足施工全流程动态可视化呈现与精准管控的迫切需求。针对上述挑战，提出一种融合四维高斯溅射(4DGS)与增强现实(AR)技术的数字孪生模型可视化交付方法，旨在实现地下空间施工过程模型的高保真动态重建与沉浸式交付。该方法构建了覆盖时序数据采集、4DGS动态建模与AR可视化交互的全链路技术框架。首先，依托多视角时序图像与三维高斯溅射(3DGS)技术，为各关键施工节点的静态模型赋予离散时间戳属性，构建四维现场施工进度的高斯溅射模型；其次，轻量化压缩3DGS模型，并部署至移动AR终端，支持不同施工阶段数字孪生体的虚实叠加对比、时序回溯、多轴向剖切与交互式实时分析。在多处典型地下工程场景的实验验证表明，经压缩的3DGS模型存储体积较传统点云和网格模型大幅缩减，且在峰值信噪比(PSNR)、结构相似性(SSIM)及感知相似度(LPIPS)等视觉保真度指标上实现显著提升。同时，该方法能在AR设备中高保真、动态清晰地还原复杂施工进程，有效规避传统模型的视觉失真问题。融合4DGS和AR驱动传统工程场景中静态、离散的模型交付模式，升级为动态、连续和可交互的沉浸式交付体验，构建形成感知、可视、交互和反馈的全闭环体系，可为地下空间工程的施工技术交底、过程管控及全生命周期运维管理提供坚实的数字底座。

关键词: 4D高斯溅射, 增强现实, 地下空间施工, 数字孪生, 施工进度可视化

Abstract:

As underground space engineering expands in scale and complexity, conventional static and discrete model delivery methods based on BIM, point clouds, or meshes suffer from slow update cycles, excessive data volumes, and visual distortions. Consequently, they are inadequate for the growing demands for dynamic visualization and precise control throughout the construction process. To address this industry challenge, a digital twin model visualization delivery framework integrating 4D Gaussian Splatting (4DGS) with Augmented Reality (AR) technology was proposed, aiming to achieve high-fidelity dynamic reconstruction and immersive delivery of underground construction models. This approach established a comprehensive technical framework encompassing temporal data acquisition, 4DGS dynamic modeling, and AR-based interactive visualization. First, by leveraging multi-view temporal imagery and 3D Gaussian Splatting (3DGS) technology, temporal attributes were integrated into the modeling of key construction nodes to construct a 4D Gaussian splatting representation of the on-site construction progress. Second, the 3DGS models were subjected to lightweight compression for deployment on mobile AR terminals. This supported virtual-real registration and overlay, temporal playback, multi-axial sectioning, and interactive real-time analysis of the digital twin across various construction phases. Experimental validation across multiple typical underground engineering scenarios demonstrated that the compressed 3DGS models achieved substantial reductions in storage footprint compared to traditional point cloud and mesh models, while significantly improving visual fidelity metrics, including Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index (SSIM), and Learned Perceptual Image Patch Similarity (LPIPS). Furthermore, this method enabled high-fidelity, dynamic, and clear reconstruction of complex construction processes on AR devices, effectively mitigating the visual distortion issues inherent in traditional models. By integrating 4DGS and AR, the conventional static and discrete model delivery paradigm was transformed into a dynamic, continuous, and interactive immersive experience, establishing a closed-loop system encompassing perception, visualization, interaction, and feedback. This provided a robust digital foundation for construction technical briefing, process control, and full-lifecycle operation and maintenance management of underground space engineering projects.

Key words: 4D Gaussian splatting, augmented reality, underground space construction, digital twin, construction progress visualization

中图分类号:

TP391
TU91

杨基好, 陈鹏禄, 陆柏桦, 赵宇, 赵春杰, 张健, 周益辉, 陈雷, 陈湘生, 谭毅. 融合4DGS和AR驱动的地下空间数字模型可视化交付方法[J]. 图学学报, 2026, 47(4): 895-903.

YANG Jihao, CHEN Penglu, LU Baihua, ZHAO Yu, ZHAO Chunjie, ZHANG Jian, ZHOU Yihui, CHEN Lei, CHEN Xiangsheng, TAN Yi. A visual delivery method for underground space digital models integrating 4D Gaussian splatting and augmented reality[J]. Journal of Graphics, 2026, 47(4): 895-903.

图/表 14

图1 时序数据采集

Fig. 1 Time-series data acquisition

图2 4DGS模型重建流程

Fig. 2 4DGS model reconstruction process

图3 AR平台架构图

Fig. 3 AR platform architecture

图4 地下空间数字模型可视化交付平台

Fig. 4 Visualized delivery platform for underground space digital model

表1 数据采集参数

Table 1 Data acquisition parameters

场景	阶段	分辨率	视频长度/s	图片数量
场景1	阶段1	3 480×2 160	76	467
	阶段2	1 920×1 080	96	582
	阶段3	1 920×1 080	112	673
场景2	阶段1	3 840×2 160	327	985
场景2	阶段2	3 840×2 160	414	1 243

图5 不同场景不同阶段的4D相机位姿估计((a) 场景1；(b) 场景2)

Fig. 5 4D camera pose estimation for different scenarios and stages ((a) Scenario 1; (b) Scenario 2)

图6 不同场景不同阶段的4D稀疏点云模型((a) 场景1；(b) 场景2)

Fig. 6 4D sparse point cloud model for different scenarios and stages ((a) Scenario 1; (b) Scenario 2)

图7 不同场景不同阶段的3DGS模型((a) 场景1；(b) 场景2)

Fig. 7 3DGS model for different scenarios and stages ((a) Scenario 1; (b) Scenario 2)

图8 同一视角下不同重建模型的细节对比

Fig. 8 Detail comparison of different reconstructed models from the same viewpoint

表2 压缩效果

Table 2 Compression effect

场景	阶段	3DGS/MB	点云/G	网格/MB
场景1	阶段1	39.5	1.55	815
	阶段2	16.5	2.73	810
	阶段3	23.4	3.80	1 966
场景2	阶段1	23.9	1.97	878
场景2	阶段2	18.4	3.24	336

表3 不同评价指标效果

Table 3 Performance of different evaluation metrics

场景	阶段	评价指标	3DGS	点云	网格
场景1	阶段1	PSNR	17.500 5	9.336 6	12.912 4
		SSIM	0.672 0	0.414 4	0.510 3
		LPIPS	0.410 6	1.054 7	0.677 3
	阶段2	PSNR	14.179 8	10.671 1	11.936 3
		SSIM	0.376 3	0.264 8	0.294 8
		LPIPS	0.376 7	0.940 4	0.620 2
	阶段3	PSNR	18.086 0	14.740 4	14.424 7
		SSIM	0.499 5	0.396 3	0.432 4
		LPIPS	0.306 3	0.948 4	0.584 3
场景2	阶段1	PSNR	20.267 1	12.596 5	17.139 5
		SSIM	0.884 4	0.409 0	0.750 9
		LPIPS	0.359 2	0.790 4	0.509 9
	阶段2	PSNR	19.086 3	12.619 2	17.311 5
		SSIM	0.876 7	0.440 9	0.748 0
		LPIPS	0.410 4	0.743 7	0.549 1

表4 不同评价指标对比

Table 4 Comparison of different evaluation metrics

场景	阶段	评价指标	3DGS vs 点云/%	3DGS vs 网格/%
场景1	阶段1	PSNR	87.44	35.53
		SSIM	62.16	31.69
		LPIPS	61.07	39.38
	阶段2	PSNR	32.88	18.80
		SSIM	42.11	27.65
		LPIPS	59.94	39.26
	阶段3	PSNR	22.70	25.38
		SSIM	26.04	15.52
		LPIPS	67.70	47.58
场景2	阶段1	PSNR	60.89	18.25
		SSIM	116.23	17.78
		LPIPS	54.55	29.55
	阶段2	PSNR	51.25	10.25
		SSIM	98.84	17.21
		LPIPS	44.82	25.26

表5 模型重建与压缩时间/min

Table 5 Model reconstruction and compression time/min

场景	阶段	SfM	重建	压缩
场景1	阶段1	523	22	6
	阶段2	676	30	3
	阶段3	783	35	4
场景2	阶段1	354	44	5
场景2	阶段2	455	49	3

图9 数字模型可视化交付现场应用((a) 4DGS模型可视化；(b) 模型多轴向剖切；(c) 信息协同标注)

Fig. 9 On-site application of visualized delivery for digital models ((a) 4DGS model visualization; (b) Multi-axial sectioning of models; (c) Information collaborative annotation)

参考文献 21

[1]	张勇, 王永明, 王春伟, 等. BIM技术在国内外项目管理中的应用现状与发展趋势[J]. 广东水利电力职业技术学院学报, 2023, 21(1): 42-46.
	ZHANG Y, WANG Y M, WANG C W, et al. Application and research trends of BIM in project management[J]. Journal of Guangdong Polytechnic of Water Resources and Electric Engineering, 2023, 21(1): 42-46 (in Chinese).
[2]	CARTER E, SAKR M, SADHU A. Augmented reality-based real-time visualization for structural modal identification[J]. Sensors, 2024, 24(5): 1609. DOI URL
[3]	杨必胜, 梁福逊, 黄荣刚. 三维激光扫描点云数据处理研究进展、挑战与趋势[J]. 测绘学报, 2017, 46(10): 1509-1516. DOI
	YANG B S, LIANG F X, HUANG R G. Progress, challenges and perspectives of 3D LiDAR point cloud processing[J]. Acta Geodaetica et Cartographica Sinica, 2017, 46(10): 1509-1516 (in Chinese). DOI
[4]	刘占省, 刘俊杰, 及炜煜, 等. 基于数字孪生的建筑工程交付模型建立及应用研究[J]. 建筑结构学报, 2024, 45(4): 97-106.
	LIU Z S, LIU J J, JI W Y, et al. Research on establishment and application of delivery model of construction engineering based on digital twin[J]. Journal of Building Structures, 2024, 45(4): 97-106 (in Chinese).
[5]	张宇, 宋伟东, 孙尚宇, 等. 基于改进3DGS的小范围场景高质量三维重建方法[J]. 激光与光电子学进展, 2026, 63(6): 0615006.
	ZHANG Y, SONG W D, SUN S Y, et al. High-quality 3D reconstruction method for small-scale scenes based on improved 3D Gaussian splatting[J]. Laser & Optoelectronics Progress, 2026, 63(6): 0615006 (in Chinese).
[6]	YU Y W, VERBREE E, VAN OOSTEROM P, et al. From comparison to integration: a workflow evaluation of 3D Gaussian splatting and LiDAR point cloud for modern architectural heritage[J]. Automation in Construction, 2025, 180: 106509. DOI URL
[7]	YU Y W, VERBREE E, VAN OOSTEROM P, et al. 3D Gaussian splatting for modern architectural heritage: integrating UAV-based data acquisition and advanced photorealistic 3D techniques[J]. AGILE: GIScience Series, 2025, 6: 51.
[8]	CHEN M L, KUANG H L, MA X L, et al. Lightweight 3D reconstruction of indoor scenes based on multi-stage gradient adjustment[C]// The 5th International Conference on Artificial Intelligence and Industrial Technology Applications. New York: IEEE Press, 2025: 1008-1011.
[9]	LI H H, SI L P. 3D Reconstruction of indoor scenes based on 3DGS models[J]. International Journal of Advanced Network, Monitoring and Controls, 2024, 9(3): 56-61. DOI URL
[10]	QIAN J T, YAN Y M, GAO F J, et al. C3DGS: compressing 3D Gaussian model for surface reconstruction of large-scale scenes based on multiview UAV images[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2025, 18: 4396-4409. DOI URL
[11]	WU C Y, HSU L S, HSU C H. FlyGS: online 3DGS model construction using a MAVLink-connected drone[C]// The 3rd International Workshop on UAVs in Multimedia: Capturing the World from A New Perspective. New York: ACM, 2025: 68-72.
[12]	杨启蒙, 傅海峰, 胡奕, 等. 复杂受限空间下“金蝉脱壳”顶管施工关键技术研究与应用[J]. 建筑施工, 2026, 48(5): 526-529.
	YANG Q M, FU H F, HU Y, et al. Key techniques and application of “Cicada-shelling” pipe jacking construction in complex and constrained space[J]. Building Construction, 2026, 48(5): 526-529 (in Chinese).
[13]	陶玉波, 蔺志刚, 董甲甲. 基于BIM的水利工程4D进度监控系统关键技术[J]. 人民黄河, 2019, 41(3): 135-139.
	TAO Y B, LIN Z G, DONG J J. 4D progress monitoring system of hydraulic project based on BIM[J]. Yellow River, 2019, 41(3): 135-139 (in Chinese).
[14]	戴成元, 梁邦勋, 程可, 等. 基于数字孪生的装配式建筑施工过程动态调度研究[J]. 建筑科学与工程学报, 2024, 41(4): 83-94.
	DAI C Y, LIANG B X, CHENG K, et al. Study on dynamic scheduling of prefabricated building construction process based on digital twin[J]. Journal of Architecture and Civil Engineering, 2024, 41(4): 83-94 (in Chinese).
[15]	周静静, 郭新宇, 吴升, 等. 基于多视角图像的植物三维重建研究进展[J]. 中国农业科技导报, 2019, 21(2): 9-18. DOI
	ZHOU J J, GUO X Y, WU S, et al. Research progress on plant three-dimensional reconstruction based on multi-view images[J]. Journal of Agricultural Science and Technology, 2019, 21(2): 9-18 (in Chinese). DOI
[16]	李鑫, 龙古灿, 刘进博, 等. 相机位姿估计的加速正交迭代算法[J]. 光学学报, 2015, 35(1): 0115004.
	LI X, LONG G C, LIU J B, et al. Accelerative orthogonal iteration algorithm for camera pose estimation[J]. Acta Optica Sinica, 2015, 35(1): 0115004 (in Chinese). DOI URL
[17]	QIAO Y H, LIAO Q X, ZHANG M R, et al. Point clouds segmentation of rapeseed siliques based on sparse-dense point clouds mapping[J]. Frontiers in Plant Science, 2023, 14: 1188286. DOI URL
[18]	刘长鑫, 刘真, 杨晟炜, 等. PSNR在网目调数字图像质量评价中的应用研究[J]. 包装工程, 2012, 33(7): 108-111, 136.
	LIU C X, LIU Z, YANG S W, et al. Application study of PSNR in quality evaluation of digital halftone image[J]. Packaging Engineering, 2012, 33(7): 108-111, 136 (in Chinese).
[19]	佟雨兵, 张其善, 祁云平. 基于PSNR与SSIM联合的图像质量评价模型[J]. 中国图象图形学报, 2006, 11(12): 1758-1763.
	TONG Y B, ZHANG Q S, QI Y P. Image quality assessing by combining PSNR with SSIM[J]. Journal of Image and Graphics, 2006, 11(12): 1758-1763 (in Chinese).
[20]	闫光辉, 覃楷超, 马龙, 等. 面向多材质光照场景的梯度自适应高斯泼溅算法[J]. 激光与光电子学进展, 2026, 63(2): 0237009.
	YAN G H, QIN K C, MA L, et al. Gradient-adaptive gaussian splatting for multi-material illuminated scenes[J]. Laser & Optoelectronics Progress, 2026, 63(2): 0237009 (in Chinese).
[21]	FISHER A, CANNIZZARO R, COCHRANE M, et al. ColMap: a memory-efficient occupancy grid mapping framework[J]. Robotics and Autonomous Systems, 2021, 142: 103755. DOI URL