Abstract: In order to guarantee the accuracy of the grounding protection wire hung by the banding robot, a research on the precise identification and positioning method of the grounding ring of the distribution network is carried out. In order to improve the recognition accuracy and easy deployment on the host computer, YOLOX-S is used as the basis, the simplified BiFPN network is used instead of the PAFPN feature utilisation layer of the neck, GIoU-Loss is used as the localisation loss function and Focal Loss is adopted as the confidence prediction loss function. A binocular vision method is used to locate the detected grounding ring, the binocular camera is calibrated to obtain the internal and external parameters and distortion coefficients of the camera, stereo correction is applied to the left and right images, the semi-global stereo matching algorithm is used to obtain the image parallax, and finally the depth and coordinates of the grounding ring are obtained using the reprojection matrix, so as to achieve accurate recognition and localization of the grounding ring. The proposed identification and positioning method is applied to the experiments of hanging grounding protection wires by a power strip operation robot, and the results show that the method can achieve accurate hanging of wires by the robot and can be extended to other operations of power robots.

Key words: target detection, YOLOX, binocular positioning, SGBM