Visual Navigation based on Deep Semantic Cues for Real-Time Autonomous Power Line Inspection

In this paper, a visual guided navigation method for Unmanned Aerial Vehicles (UAVs) during power line inspections is proposed. Our method utilizes a deep learning-based image segmentation algorithm to extract semantic masks of the power lines from onboard camera images. These masks are then processed and visual characteristics along with geometrical calculations generate velocity commands for the 3D position and yaw control that feed the UAV’s navigation system. The accuracy, robustness, and computational efficiency of the power line segmentation module are evaluated on real benchmark datasets. Extensive simulation experiments have been conducted to assess the proposed method’s performance in terms of inspection coverage, considering various textured environments and extreme initial states. The proposed method for navigating a UAV towards target PTLs is shown to be effective in terms of robustness and stability. This is achieved through accurate segmentation of the PTLs and the generation of compact velocity directives based on visual information in various environmental conditions. The results indicate a significant improvement in the precision of autonomous UAV-based inspections of power infrastructure due to continuous scoping of the transmission lines and safe yet stable navigation.