Utility-Pole, Nuts and Cross-Arm Visual Detection, for Electric Connections Maintenance Robot

Abstract

In this work, a computer vision system used by a utility-pole maintenance robot is presented. The objective of the system is that the robot repairs connections between cables and screws on the utility-pole. To accomplish this task, it uses a pair of cameras that detects and locates the utility pole, crosstree, nuts, and screws. The detection is based on color, shape and dimensions analysis, by using conventional RGB cameras. On the other hand, the challenges facing this approach include: utility-pole detection though in the same scene, some objects with the same color (concrete) appear; flexibility in the distance between cameras and utility-pole for detection and localization, about 2-10 meters range and slope estimation of utility-pole and crosstree. The experiments show that it is possible to recognize and locate the interest pieces with satisfactory results and they encourage to expand the current system capabilities.