面向电力设备的三维重建技术在状态感知、故障诊断、数字孪生和智能巡检等工作中具有重要意义 。针对 热红外图像纹理弱、对比度低、细节不足，导致传统多视图重建中特征匹配困难、位姿估计不稳定和模型不完整等 问题，文中提出一种基于视觉几何组运动恢复结构（visual geometry grounded deep structure from motion，VGGSFM）与 改进三维高斯溅射（3d gaussian splatting，3DGS）的三维重建方法 。首先，利用VGGSFM对多视角热红外图像进行相 机位姿估计和稀疏结构恢复，为后续重建提供几何先验；随后，以相机参数和稀疏点云初始化3DGS显式场景表示， 实现热红外场景重建与新视角表达；最后，引入深度法向正则约束，增强高斯元分布的几何稳定性和结构一致性 。
[1]胡钰泊，李厚君，游桂杨.基于YOLOv8的电力设备红外图 像检测[J].电气技术与经济，2026(01):222-226.
[2]CHIH,LUO DL,WANG S.LMDFusion:A lightweight in- frared and visible image fusion network for substation equip- ment based on mask and residual dense connection[J].Infra- red Physics &Technology,2024,138:105218.
[3]张秋铭，李云红，罗雪敏，等.改进Chan-Vese模型的电力设 备红外图像分割算法[J].红外技术，2023,45(02):129-136.
[4]SkorokhodovV,Xu C,Sun S,et al.SEAR:Simple and efficient adaptation ofvisual geometric transformers for RGB+Thermal 3D reconstruction[EB/OL].arXiv,2026:arXiv:2603.18774.
[5]VASWANI A,SHAZEERN,PARMARN,et al.Attention is all you need[C]//Advances in Neural Information Processing Systems.Red Hook,NY:Curran Associates,Inc.,2017,30: 5998-6008.
[6]SENTENACT,BUGARIN F,DUCAROUGE B,et al.Auto- mated thermal 3D reconstruction based on a robot equipped with uncalibrated infrared stereovision cameras[J].Advanced Engineering Informatics,2018,38:203-215.
[7]KERBLB,KOPANAS G,LEIMKUHLER T,et al.3D gauss- ian splatting for Real-Time radiance field rendering[J].ACM Transactions on Graphics,2023,42(04):1-14.
[8]WANG JY,KARAEVN,RUPPRECHTC,et al.VGGSfM:Vi- sual geometry grounded deep structure from motion[C].Pro- ceedings ofthe IEEE/CVF Conference on Computer Vision and Pattern Recognition.Seattle,WA:IEEE,2024:21686-21697
