Feature Points Extraction Based on Line Fitting for Power Distribution Switch

During inspection and operation tasks in the power distribution room, it is necessary for unattended robots to have the ability to acquire high-precision position and attitude information of each distribution switch. For the common square distribution switch, this paper proposes a highly versatile and high-precision feature points extraction method, which can be used for positioning the switch in three-dimensional space. This method combines the classification advantages of machine learning with the positioning advantages of traditional image algorithms, and obtains feature points for matching with a series of methods such as Hough transform and least squares. The matching feature points obtained by the algorithm are fitted by a large number of edge contour points, thus offering high precision and high robustness. The experiment shows that the precision of calculated switch position based on our method can reach 1.5 mm in translation and 1.5 degrees in rotation, which meets the requirements of unattended robot operation.