Abstract:

In view of the problem of low detection efficiency of high-speed train pantograph in the complex environment of passing through the support bridge tunnel, a dynamic monitoring method of the pantograph-net contact position under complex environment is proposed. In order to obtain the original training data set, the pantograph video is captured by frame difference. The deep learning network PSPNet is used to semantically segment the contact line and the pantograph of the image, which is used to construct the feature data set with more obvious contact points of pantograph. To obtain the coordinates, the improved YOLOv4 is used for training and detection. The results show that the proposed method can effectively mark the contact point position between pantograph and catenary in each frame image, and can capture the movement state of pantograph and output the relative coordinate position when the train passes through the support frame and bridge, so as to achieve the monitoring purpose of pantograph, and the detection accuracy of the proposed method can reach 96.8%.

Key words: pantograph monitoring, feature extraction, feature data set, semantic segmentation, objection tracking, deep learning, PSPNet, YOLOv4