Research On Pantograph Catenary Contact Point Localization Technology Based On Multi-stage Detection

At present,the rapid development of urban rail transit in China,the normal operation of traction power supply system for improving the safety of rail transit system operation is of great significance.Among them,pantograph and catenary are important components of traction power supply system.Studying the position of the contact point of pantograph is an important step for the subsequent parameter calculation of pantograph and the normal power supply monitoring of trains.This paper mainly focuses on the detection of contact points between rigid catenary and pantograph in urban rail transit,and this paper also studies the location method of contact points of pantograph based on multi-stage detection.Firstly,the pantograph video images were extracted,and the initial pantograph images were obtained through frame segmentation and preliminary screening of the collected data.The pantograph contact areas and key points of pantograph were marked by image annotation tools,and experimental data sets were generated by digital image processing technology.Due to the relatively humid tunnel environment,the camera lens fogged and the image effect was not good.Therefore,the image fogging method was further studied.Through comparison,it is found that compared with the histogram equalization algorithm,the dark channel prior algorithm has a more ideal effect in defogging.After image brightness compensation,the defogging image is obtained,which lays a data foundation for subsequent image detection.In the first stage,the detection network of the arch network contact area was designed based on the YOLOv4 network,and the Mosaic image data enhancement method was studied.The Kmeans clustering algorithm was used to calculate the priori box data,and the CIo U loss function was used to calculate the loss.The experimental comparison was made between the detection accuracy,iteration speed and prediction time and the YOLOv3 network model.In the second stage,the contact point detection network is designed based on stacked hourglass network,and the MISH activation function and RMSE loss function were used to improve the positioning network.Four key points generated by the intersection of rigid catenary and pantograph in the image are determined.The influence of image defogging treatment on the experimental results was analyzed and an adaptive detection scheme for key points was designed.Finally,using the mathematical model of multiple linear regression,the contact point coordinates of the pantograph are located and mapped from the contact area coordinates to the original image coordinates.The experimental results show that,no matter whether the pantograph image is interfered by fog or the pantograph has multiple contact areas in the switching stage of catenary,this experimental method can obtain relatively accurate location of the contact area and detection of the contact points of the pantograph.Among them,the target detection method based on the YOLOv4 network in the first stage of this paper is more accurate than that based on the YOLOv3 network.In the second stage of this paper,the detection method of pantograph contact points based on stacked hourglass network can accurately locate the position of pantograph contact points under the condition of fog interference.This method can be used as a reference for other kinds of contact point detection between catenary and pantograph.