Invasion detection on transmission lines using saliency computation

This paper proposes a method of invasion detection on transmission lines using saliency computation. Previous researches mainly extract transmission lines with Canny edge detector and Hough transformation (HT), then detect invasions from the topological characteristics along the lines. However, HT works poorly in the system where transmission lines are curves, and too many false detection will arise when background is complex. Moreover, topological methods cannot perform well when invasion objects are long and narrow with little difference from lines in color and shape. The proposed method focuses on improving the precision of transmission line extraction and integrates saliency detection into the procedure of invasion object detection. We adopted statistical color filtering to line segment detector, which is tailored to fit curves in pixel level and match the color characteristics of transmission lines. In addition, we use saliency computation to uniformly extract the foreign matter on transmission lines since these invasions always appear as pop-up regions in the image sequences. Comparative experiments are conducted to evaluate the performance of the proposed method with two previous works, which indicate that our work can achieve at least 30 percent higher precision in the worst cases than the other two methods and take 40ms~200ms less time for processing each image frame.