Research on Gaussian Splatting Modeling of Power Equipment in 3D Scenes

doi:10.16182/j.issn1004731x.joss.25-0628

Abstract

Abstract:

To address the issues of missing camera poses in captured images and poor reconstruction quality in 3D modeling of power equipment, a 3D Gaussian splatting 3D modeling method for power equipment based on video sequences wasproposed. Theffmpeg was adopted to extract video frames at a reduced rate, and the Scharr operator was employed to quantify the sharpness of video frames to screen high-quality images for forming an input dataset, ensuring the completeness of equipment poses and the quality of modeling data. Through multi-view feature point extraction and matching, combined with an incremental structure-from-motion algorithm to optimize and generate a sparse 3D point cloud, the geometric foundation of the model was established. 3D Gaussian point clouds constructed from the sparse point cloud were projected onto the image plane using Gaussian splatting, and Gaussian parameters were iteratively optimized by designing a loss function.Differentiable rasterization rendering technology was integrated to generate a photorealistic 3D Gaussian model of power equipment. Experimental results of power equipment modeling indicate that the proposed method can efficiently reconstruct 3D models with rich details and accurate geometryand can form three-dimensional global visualization in virtual space by combining multi-modal monitoring data, possessing important engineering application value.

Key words: power equipment, sharpness score, structure-from-motion, Gaussian splatting, 3D reconstruction

CLC Number:

TM93

Li Haiying, Xu haonan, Hao Junfang. Research on Gaussian Splatting Modeling of Power Equipment in 3D Scenes[J]. Journal of System Simulation, 2026, 38(3): 595-607.

Figures/Tables 22

Fig. 1

Fig. 2

Fig. 3

Fig. 4

Fig. 5

Fig. 6

Fig. 7

Fig. 8

Fig. 9

Fig. 10

Fig. 11

Fig. 12

Fig. 13

Fig. 14

Fig. 15

Fig. 16

Table 1

Fig. 17

Table 2

Fig. 18

Fig. 19

Table 3

References 28

[1]	中国南方电网. 数字电网推动构建以新能源为主体的新型电力系统白皮书[R]. 广州: 中国南方电网有限责任公司, 2021.China Southern Power Grid Company Limited. Digital Power Grid Promotes the Construction of a New Type Power System with New Energy as the Main Body[R]. Guangzhou, China: China Southern Power Grid Company Limited, 2021.
[2]	国家电网有限公司. 新型电力系统数字化支撑技术体系白皮书[R]. 北京: 国家电网有限公司, 2022.State Grid Corporation of China.The Digital Support Technology System of New Type Power Systems[R]. Beijing: State Grid Corporation of China, 2022.
[3]	赵辉, 韩璟琳, 李光毅, 等. 基于运动恢复结构原理的电网设施三维静态重建方法[J]. 现代电力, 2024, 41(6): 1023-1031.
	Zhao Hui, Han Jinglin, Li Guangyi, et al. 3D Static Reconstruction Method of Power Grid Facilities Based on Principle of Structure from Motion[J]. Modern Electric Power, 2024, 41(6): 1023-1031.
[4]	赵鹏, 蒲天骄, 王新迎, 等. 面向能源互联网数字孪生的电力物联网关键技术及展望[J]. 中国电机工程学报, 2022, 42(2): 447-457.
	Zhao Peng, PuTianjiao, Wang Xinying, et al. Key Technologies and Perspectives of Power Internet of Things Facing with Digital Twins of the Energy Internet[J]. Proceedings of the CSEE, 2022, 42(2): 447-457.
[5]	江伦, 王大江, 孙文磊, 等. 基于数字孪生的变压器故障诊断方法研究[J]. 系统仿真学报, 2025, 37(3): 775-790.
	Jiang Lun, Wang Dajiang, Sun Wenlei, et al. Research on Transformer Fault Diagnosis Method Based on Digital Twin[J]. Journal of System Simulation, 2025, 37(3): 775-790.
[6]	沈小军, 于忻乐, 王远东, 等. 变电站电力设备红外热像测温数据三维可视化方案[J]. 高电压技术, 2021, 47(2): 387-395.
	ShenXiaojun, Yu Xinle, Wang Yuandong, et al. Three-dimensional Visualization Scheme of Infrared Thermal Temperature Measurement Data for Substaition Electric Power Equipment[J]. High Voltage Engineering, 2021, 47(2): 387-395.
[7]	董少迪, 唐兆, 王开云, 等. 数字孪生在列车曲线通过性能预测中的应用研究[J]. 机械工程学报, 2022, 58(18): 240-250.
	Dong Shaodi, Tang Zhao, Wang Kaiyun, et al. Application of Digital Twin to Curve Negotiation Performance Prediction of Train[J]. Journal of Mechanical Engineering, 2022, 58(18): 240-250.
[8]	Cui Yunge, Zhang Yinlong, Dong Jiahua, et al. LinK3D: Linear Keypoints Representation for 3D LiDAR Point Cloud[J]. IEEE Robotics and Automation Letters, 2024, 9(3): 2128-2135.
[9]	陆后军, 朱益飞, 荣延平, 等. 基于激光雷达的散货堆场数字孪生建模方法[J]. 系统仿真学报, 2025, 37(9): 2269-2286.
	Lu Houjun, Zhu Yifei, RongYanping, et al. Digital Twin Modeling Method for Bulk Cargo Stacks Based on 2D LiDAR[J]. Journal of System Simulation, 2025, 37(9): 2269-2286.
[10]	徐召飞, 廖键, 王宏臣, 等. 面向道路场景的红外与激光雷达配准算法研究[J]. 激光与光电子学进展, 2024, 61(14): 302-312.
	XuZhaofei, Liao Jian, Wang Hongchen, et al. Research on Infrared and LiDAR Calibration Algorithm for Road Scene[J]. Laser & Optoelectronics Progress, 2024, 61(14): 302-312.
[11]	Cernea D. OpenMVS: Multi-view Stereo Reconstruction Library [DB/OL]. [2024-09-23]. .
[12]	姜骞, 刘亚东, 严英杰, 等. 电力设备多源异构数据空间合成与立体展示方法[J]. 高电压技术, 2022, 48(1): 66-74.
	Jiang Qian, Liu Yadong, Yan Yingjie, et al. Spatial Synthesis and Stereoscopic Display Method for Multi-source Heterogeneous Data of Power Equipment[J]. High Voltage Engineering, 2022, 48(1): 66-74.
[13]	Mao Yongqiang, Bi Hanbo, XuLiangyu, et al. SDL-MVS: View Space and Depth Deformable Learning Paradigm for Multiview Stereo Reconstruction in Remote Sensing[J]. IEEE Transactions on Geoscience and Remote Sensing, 2024, 62: 1-18.
[14]	Mildenhall B, Srinivasan P P, Tancik M, et al. NeRF: Representing Scenes as Neural Radiance Fields for View Synthesis[J]. Communications of the ACM, 2021, 65(1): 99-106.
[15]	Barron J T, Mildenhall B, Tancik M, et al. Mip-NeRF: A Multiscale Representation for Anti-aliasing Neural Radiance Fields[C]//2021 IEEE/CVF International Conference on Computer Vision (ICCV). Piscataway: IEEE, 2021: 5835-5844.
[16]	Hedman P, Srinivasan P P, Mildenhall B, et al. Baking Neural Radiance Fields for Real-time View Synthesis[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025, 47(5): 3310-3321.
[17]	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.
[18]	Pei Shaotong, Yang Rui, Liu Yunpeng, et al. Research on 3D Reconstruction Technology of Large‐scale Substation Equipment Based on NeRF[J]. IET Science, Measurement & Technology, 2023, 17(2): 71-83.
[19]	Kerbl Bernhard, GeorgiosKopanas, Leimkuehler Thomas, et al. 3D Gaussian Splatting for Real-time Radiance Field Rendering[J]. ACM Transactions on Graphics, 2023, 42(4): 139.
[20]	Yang Ziyi, GaoXinyu, 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.
[21]	HelgeRhodin, Robertini Nadia, Richardt Christian, et al. A Versatile Scene Model with Differentiable Visibility Applied to Generative Pose Estimation[C]//2015 IEEE International Conference on Computer Vision (ICCV). Piscataway: IEEE, 2015: 765-773.
[22]	Scharr H. Optimal Operators in Digital Image Processing[D]. Heidelberg: Heidelberg University, 2000.
[23]	Yin Huilin, Sun Mina, Zhang Linchuan, et al. Online Dynamic Object Removal for LiDAR-inertial SLAM Via Region-wise Pseudo Occupancy and Two-stage Scan-to-map Optimization[J]. Displays, 2025, 88: 103030.
[24]	宁小娟, 李洁茹, 高凡, 等. 基于最佳几何约束和RANSAC的特征匹配算法[J]. 系统仿真学报, 2022, 34(4): 727-734.
	NingXiaojuan, Li Jieru, Gao Fan, et al. Feature Matching Algorithm Based on Optimal Geometric Constraints and RANSAC[J]. Journal of System Simulation, 2022, 34(4): 727-734.
[25]	江晓斌. 基于无人机影像的增量式SFM关键技术研究[D]. 淄博: 山东理工大学, 2020.
	Jiang Xiaobin. Research on Key Technologies of Incremental SFM Based on UAV Image[D]. Zibo: Shandong University of Technology, 2020.
[26]	Szeliski R. 计算机视觉: 算法与应用[M]. 北京: 清华大学出版社, 2012.SzeliskiR. Computer Vision: Algorithms and Applications[M]. Beijing: Tsinghua University Press, 2012.
[27]	Hahlbohm Florian, Friederichs Fabian, Weyrich Tim, et al. Efficient Perspective-correct 3D Gaussian Splatting Using Hybrid Transparency[J]. Computer Graphics Forum, 2025, 44(2): e70014.
[28]	Jiang Qian, Liu Yadong, Yan Yingjie, et al. BIM-based 3D Multimodal Reconstruction for Substation Equipment Inspection Images[J]. IEEE Transactions on Instrumentation and Measurement, 2024, 73: 1-14.

指标	筛选前	筛选后
模型迭代时间/min	33.93	31.86
PSNR/dB	28.68	29.63
SSIM	0.867	0.892

迭代次数	PSNR/dB	SSIM	Loss	迭代时间
6 000	24.129	0.847	0.064	5 min35 s
12 000	26.395	0.859	0.051	11 min27 s
18 000	28.136	0.873	0.047	18 min12 s
24 000	28.928	0.886	0.042	25 min03 s
30 000	29.632	0.892	0.035	31 min51 s

算法名称	构建时间	PSNR/dB	SSIM
SfM+MVS	78 min	—	—
Instant ngp	10 min	25.86	0.832
Mip-NeRF	48 h	29.83	0.887
3DGS①	10 min	26.10	0.853
3DGS②	31.86 min	29.63	0.892