Detection for overhead ground wire by lines clustering

Abstract

In order that such adverse factors as complex background, low contrast between the background and overhead ground wire and blurred edges of the wire in on-site environment of hand-eye vision can be overcome, an object segmentation method based on straight lines clustering (SLC) is designed by improving the segmentation algorithm based on pixel clustering, so that pose parameters of ground wire can be detected from images. After edge detection of the images is completed, the edges are divided into straight lines with certain straightness. The lines are detected and such data as normal angles θ, distances d to the image origin and the endpoint coordinates are acquired. Then the parameter pairs (d, θ) are clustered as points on plane dθ, and the lines with similar parameter pairs are classified into the same category. For each category, outer lines are eliminated, and the remaining lines are used to determine its segmentation region, at the same time, the pose parameters of it are obtained. For the segmented objects, their local binary pattern features are extracted and SVM classifier is used to determine whether they are ground wire. The test result shows that the detection accuracy of the pose parameters of the wire can reach 90.8% and the proposed method is effective.