Abstract: Abstract : The digitization of grid equipment is a key step of power grid digital twinning. The traditional manual modeling method based on design data and ledger data is inefficient and far from reality. To address the above drawbacks, a 3D static model reconstruction method for digital twin power grid facilities was studied, considering the difficulty in extracting feature points due to the weak texture and single color of some power grid facilities, as well as the real-time and accuracy demands of 3D reconstruction. Firstly, based on the Structure from motion (abbr. SFM), a complete algorithm flow design was carried out for the steps of pose estimation, dense point cloud extraction, and surface reconstruction in this principle. Secondly, the accuracy of camera pose estimation was improved by improving the matching conditions of feature points and combining the five-spot method with the random sample consensus algorithm. Thirdly, the minimum eigenvalue algorithm and the Kanade-Lucas-Tomasi(abbr. KLT) optical flow method were combined to increase the selection number of dense point clouds, and ultimately effectively improve the accuracy and visual effect of point cloud reconstruction. Finally, the combined transformer box was taken as an example for experimental verification. The experimental results and precision comparison analysis show that the proposed method can quickly and accurately construct grid facilities’ virtual model, which effectively supports the digital transformation of the traditional power grid and the implementation of a digital twin power grid.