Vision-based detection of power distribution lines in complex remote surroundings

Inspection and Maintenance of power lines is an important and costly legal responsibility, mainly for public safety, of a power distribution company. The traditional method of manual, on-foot inspections have many disadvantages such as long inspection cycles etc. Due to advances in sensors and flight technology, an emerging cost-effective solution is to employ Unmanned Aerial Vehicles (UAV) for the inspections. Automatic detection and complete extraction of power lines from UAV-based aerial imagery, having complex and varying natural surroundings, is first critical problem to be efficiently solved for enabling detection of line faults. In this paper, we propose a solution based on a novel morphological operator, and a robust image space heuristics for locating and complete extraction of power lines. Our algorithm is a three-stage algorithm, and it focuses on minimization of missed detection of line segments. The entire algorithm was tested on a real outdoor video shot for around 320 meters length of power grid using a fixed-wing UAV. No missed detection of important line segments across the sequence of overhead video frames, and <; 3% false positives prove that our approach is very effective. We believe that our algorithm can be easily ported for line detection problem in any other real outdoor video as well.