Gradient Based Power Line Insulator Detection

Abstract

Insulators are widely used in the power system to provide electrical insulation and mechanical support for high voltage transmission lines. Detecting and localizing the insulators automatically are very important to intelligent inspection, which are the prerequisites for fault diagnose. A method for insulator detection in the image of overhead transmission lines based on Histogram Oriented Gradient(HOG) is presented in this paper. Dividing our work into two phases- electrical pole detection and power line insulator detection. In the first context, present a line based approach for automated detection of electrical poles. The detection is performed by first applying edge detection algorithm. Then, Applying Probabilistic Hough Transform to extract line information from the image. We Prepare data by considering only vertical lines and a variation in angle of ±5-degree as an empirical value. Group the pre-processed data and find the coverage area by applying a Heuristic function. Take the best three coverage group which consists of lines that create three individual lines. In those best three lines at least one line detects a pole which maximize the coverage area. After detecting electrical pole, we go to the top of the pole and take a zoom in picture where we search for insulator. Before test the image we take different insulator and non-insulator images as our training Dataset. To train a classifier using support vector machine in its LIBSVM tools, extract HOG features of training Dataset. As classifier have been trained we apply Sliding-window object detection technique for identifying and localizing insulators in an image. The approach involves scanning the image with a fixed-size rectangular window. Extract HOG features of the sub-image defined by the window and apply classifier to check that the window bounds an insulator or not. The process is repeated on successively scaled copies of the image so that objects can be detected at any size. To detect angled insulator, rotate the image into 360-degree and apply sliding window technique with image pyramids to detect insulators at varying scales and locations in the image. Sliding window technique detects some non insulator images as an insulator called false positive. To remove false positive Hard Negative Mining is applied and re-train classifier using those false positive samples which improve the performance of the classifier from an initial run of the classifier. Usually nonmaximal suppression is applied to the output to remove multiple detections of the same insulator. Experimental results showed that the method could extract the insulator from the image precisely, and it was suitable for many practical applications such as insulator fault diagnosis, insulator contamination grade determination and so on. Compare to the other Features like -LBP, experimental results indicate that HOG based feature detect insulator more effectively and accurately and detect insulators in different angle under complex background.