湖南电力 ›› 2025, Vol. 45 ›› Issue (2): 110-115.doi: 10.3969/j.issn.1008-0198.2025.02.015

• 配电网与用能技术 • 上一篇    下一篇

基于结构感知三维高斯泼溅技术的变电站三维模型重建方法

康文, 丁玉柱, 邓维, 钟昱铭, 唐江琦   

  1. 国网湖南省电力有限公司超高压变电公司,湖南 长沙 410004
  • 收稿日期:2024-10-29 修回日期:2024-12-23 发布日期:2025-04-30
  • 通信作者: 丁玉柱(1986),男,高级工程师,研究方向为电力设备带电检测。
  • 作者简介:康文(1982),男,高级工程师,主要研究方向为超、特高压输电智能运检技术等。
  • 基金资助:
    国网湖南省电力有限公司实验室专项(5216A324000B)

A Three-Dimensional Model Reconstruction Method of Sub‍station Based on Structure-Aware 3D Gaussian Splatting

KANG Wen, DING Yuzhu, DENG Wei, ZHONG Yuming, TANG Jiangqi   

  1. State Grid Hunan Extra High Voltage Substation Company, Changsha 410004, China
  • Received:2024-10-29 Revised:2024-12-23 Published:2025-04-30

摘要: 针对变电站三维模型重建目前使用的点云补全技术或三维扫描技术均依赖于高精度的激光雷达硬件,只能重建稀疏的、外观不精确的三维模型的问题,提出基于可见光图像的双分支三维模型重建方法,极大地减少了对硬件设备的依赖,通过互几何监督和双向引导充分发挥符号距离场和三维高斯泼溅技术的优势,高效地重建纹理丰富、稠密、高保真的变电站场景及其设备模型。

关键词: 变电站, 三维模型重建, 三维高斯泼溅

Abstract: Aiming at the problem that the point cloud complementation technologies or three-dimensional (3D) scanning techniques currently used in 3D model reconstruction of substations rely on high-precision lidar hardware and can only reconstruct moderately sparse and visually inaccurate 3D models, the dual-branch 3D modeling network based on visible light images is proposed, which greatly reduces the dependency on hardware equipment. By fully exploiting the advantages of signed distance fields and 3D Gaussian splatting through mutual geometric supervision and bidirectional guidance, it can efficiently reconstruct texture-rich, dense, and visually high-fidelity substation scenes and their equipment models.

Key words: transformer substation, 3D model reconstruction, 3D Gaussian splatting

中图分类号: