A Method Combining Line Detection and Semantic Segmentation for Power Line Extraction from Unmanned Aerial Vehicle Images

Power line extraction is the basic task of power inspection with unmanned aerial vehicle (UAV) images. However, due to complex backgrounds and limited characteristics, from images a difficult problem. In this paper, we construct data set using UAV classify according image clutter (IC). A method combining detection semantic segmentation used. This divided into three steps: First, multi-scale LSD used determine candidate regions. Then, based on object-based Markov random field (OMRF), weighted region adjacency graph (WRAG) constructed distance angle information segments capture interaction between objects, which introduced Gibbs joint distribution label field. Meanwhile, Gaussian mixture model utilized form likelihood function by taking spectral texture features. Finally, Kalman filter (KF) least-squares are realize pixel tracking fitting. Experiments carried out test in set. Compared common methods, proposed algorithm shows better performance different IC. study can provide help guidance for inspection.