Domain Adaptation in Power Line Segmentation: A New Synthetic Dataset

The task of autonomous inspection of the power transmission infrastructure includes a robotic system that is able to track the power line. This work addresses the problem of the visual localization of the power lines from RGB cameras, using deep neural networks. For challenging-to-label tasks like this, simulators can efficiently generate large amounts of labeled data. In this work, a large-scale annotated synthetic power lines dataset has been generated\footnote{That will be made publicly available upon paper acceptance.}. To address domain shift between real and synthetic domain, input-level adaptation was performed. Additionally, a new power line segmentation loss developed to mitigate the effects of unbalanced pixel distributions among power lines and background. Experiments demonstrate that our approach achieves state-of-the-art performance on power line segmentation task.