Abstract:

The conditions under which pictures are taken incur varied background complexity in aerial images. How to maintain consistent detection results in diverse backgrounds becomes an important issue. In this paper we propose an adaptive method for insulator blew detection. It locates the insulator with a 3D model and detects defects using the template of the insulator sheet. Firstly, we match the 3D model with standard image by the sift feature and find the transformation matrix from the standard image to the image to be tested via the RANSAC algorithm so as to locate all the insulators. Secondly, we transform the image from RGB to HSV, deal the H-channel with wavelet transform, and make it a binary image. We determine the direction of the axis in the binary image and turn it vertical by the Radon transform. After noise removed, we take the first insulator sheet as a model, which slides over the axis to record the histogram of every position. We employ the Earth Mover’s Distance (EMD) method to judge whether the histogram is normal or not, and  locate the self-destruction position. Experimental results indicate the proposal is highly accurate and robust.

Key words: 3D model, image matching, RANSC, HSV model, sliding model