A Coarse-to-Fine Detection Method of Pantograph-Catenary Contact Points Using DCNNs

Abstract The pantograph catenary system is an important part of the traction power supply system. In order to monitor the dynamic parameters of the rigid catenary accurately in real-time, we propose a new coarse-to-fine approach to locate and detect the pantograph-catenary contact points using DCNNs. For the contact area, which is relatively small enough for the target region, we refer to a real-time object detection method with both speed and accuracy advantages, YOLOv3. And then based on the geometric relationship between the pantograph and the catenary, we use Hoff line detection to achieve accurate detection of contact points. Our method consists of two stages. We first train Yolov3 to detect the local region of the contact points accurately by using the pantograph-catenary datasets. Obtained images of the coarse region detection, we then choose the canny edge detection and Hough transformation to detection the pantograph-catenary contact points. The experiment results from two video datasets show that our proposed method can accurately track the pantograph-catenary contact points by the continuous detection, which could provide research refers to the real-time automatic monitoring of the pantograph-catenary system.