Autonomous monitoring system for the detection of disrupting power cables on distribution networks using computer vision techniques

Abstract

This paper aims on evaluating classical computer vision techniques in order to detect whether electric cables are falling to the ground or not. A proprietary dataset containing several simulation cases was built with the Unreal Engine 4 framework and simulations were performed on MATLAB with an algorithm that performs frame subtraction and applies filters in order to detect movement on images. Evaluated metrics were precision, recall, F1 score and Frames Per Second (FPS) processed. The algorithm performed satisfactorily on most cases, being able to achieve a precision of up to 100.00%, 26.67% recall, 0.4058 F1 score and up to11 FPS.