Abnormality Detection of Ground Wire Based on Color Histogram using Images Taken from Monitoring Machine

Abstract

Arc marks and cut wires on an ground wire are mainly checked through by a helicopter. When the helicopter cannot be used, a machine that incorporate a video camera is used. The machine attached wheels runs on the ground wire and takes a video of ground wire. After recoding videos, a worker check whether or not, there is an arc mark and cut wire in the video. There are few faults in the video. The task is very bored for the worker, therefore, it is required to reduce the amount of the video that the worker has to check. We have developed a new method that extracts images that could include those faults and discards other images. The method detects an arc mark, cut wire and corrosion product that appears on the surface of the ground wire due to inner corrosion, based on color feature histogram. The features are learned by one of machine learning method, which is called Support Kernel Machine (SKM). To verify the method, 100 images including arc marks and 186 images including corrosion products are used. 89 arc marks images are detected, 169 images that corrosion products appear are detected. Through the verification, the effectiveness of the proposed method was presented.