An Early Warning Method of Pantograph Horn Drilling Based on Superpixel HOG Algorithm and YOLO v3 smart Detector

EasyChair Preprint 1526

Abstract

To slove the contact net drilling problem caused by horn intrusion during pantograph operation, a warning method for the target of pantograph arch is proposed. Firstly, the method performs the target feature area on the received pantograph images by clustering algorithm to realize the target image feature extraction and the pantograph outline feature extraction, and through the clustered contour feature to obtain the location of the largest contact point area of the pantograph and catenary. Then, the target feature area is segmented from the background by superpixel HOG target segmentation algorithm, and the target data set is thus formed by the labeled maximum feature images. Finally, The YOLOv3-smart detector is adopted to build classification model. The results showed that the proposed method could accurately track and extract the contact area of the pantograph and the catenary from the video, and had an effective significance for early warning of the pantograph drilling problem.

Keyphrases: Early Warning Method, Superpixel HOG, YOLOv3-smart, feature extraction, pantograph

@booklet{EasyChair:1526,
  author    = {Enhong Wang and Shubin Zheng and Qianwen Zhong and Liming Li and Qiaomu Zhang},
  title     = {An Early Warning Method of Pantograph Horn Drilling Based on Superpixel HOG Algorithm and YOLO v3 smart Detector},
  howpublished = {EasyChair Preprint 1526},
  year      = {EasyChair, 2019}}