Optical Detection and Altitude Estimation of Power Lines Relative to a UAV

Abstract: Inspection of power lines is a crucial activity to locate the source of power outages or to preemptively stop them. Visual inspection methods using optical cameras attached to Unmanned Aerial Vehicles (UAV) have become increasingly popular due to cheap operating costs, large areal coverage and the wide availability of these platforms. The efficiency of power line inspection can further be improved using an automated UAV system, though the UAV must maintain a safe distance from the power lines to reduce the risk of collision. This thesis proposes a method of detecting visually straight power line cables and estimating the position of a UAV with respect to the detections in real time. Two power line detection strategies are investigated; a Canny and a Laplacian edge detection based approach. The algorithms are evaluated on power lines images in a range of lighting conditions, backgrounds and transmission cables. The Canny approach is chosen as the line detection method due to a superior performance when evaluated with a recall and precision line detection metric. Using this approach, UAV translational information with respect to the power lines is calculated by using the distance between the outermost power line cables and UAV odometry measurements. A complete evaluation of the algorithm is performed in both a simulated environment and on recordings of real power lines. The UAV translational error relative to the power lines was found to be below 0.1 m in general when using a simulated environment. Using real data, where only the altitude could be evaluated, the proposed method achieved an error of 1.6 m. The greater error was caused by inaccurate power line detections in conditions with poor cable visibility and uncertain ground truth values due to terrain approximations. Overall, the proposed algorithm provides a UAV translation estimate with respect to transmission lines in real time when individual cables are easily distinguishable from the background, thus improving the capabilities of automated power line inspection.