It is fundamental to ensure the safety of distribution substations that the operation and maintenance personnel complete the standardized actions in the specified position. In this paper, a method of three-dimensional (3D) tracking and positioning the operation and maintenance personnel is proposed by collecting two-dimensional (2D) images, extracting 2D joint data, and fusing them to produce 3D joint data. Firstly, the calibration target is used to calibrate the multiple cameras installed in the distribution substation. Secondly, the YOLO-Pose model is adopted to extract the joint coordinates of multiple human 2D skeletons. Thirdly, the matching of different operation and maintenance personnel in different cameras is realized based on the epipolar constraint relationship of binocular cameras. The combination of the distance error obtained by the epipolar constraint relationship and the confidence level of 2D joint coordinates is used for the selection of two cameras to reconstruct 3D joint points of operation and maintenance personnel based on the principle of multiple-view stereo. The study will lay a foundation for the behavior identification and evaluation, behavior inversion of operation and maintenance personnel, and the establishment of electronic fences in distribution substations.