Metodologia baseada em visão computacional para rastreamento e pouso em linhas de transmissão com VANTs

Abstract

A power transmission system must have, among other requirements, high levels of availability so that the country’s energy resources can be optimally exploited. One way to ensure this is through preventive actions, such as inspection and maintenance routines. In this sense, the use of Unmanned Aerial Vehicles (UAVs) proves to be an efficient alternative, with good performance and cost-effectiveness. The development of embedded systems for these vehicles, supported by computer vision techniques, for example, can make inspection tasks more robust, safe, and agile, reducing or even eliminating the necessary manual labor. Therefore, this work proposes a system based on computer vision algorithms and models to assist in tracking and landing tasks on transmission lines, the latter in conjunction with a Human-in-the-loop methodology. The results presented, obtained through simulations in virtual scenarios, demonstrate that the proposed system is capable of correctly identifying transmission lines in aerial images captured by the vehicle, extracting the information of interest, and successfully executing the desired tasks. It was possible to develop an object detection model with mAP of 98.47% and reach the desired points in flight missions with small errors, in the order of centimeters. For all cases, the level of human interference required was minimized, making the entire process less dependent on the operator’s expertise in UAV control.