Research On Abnormal State Detection System Of High-speed Rail Pantograph Based On Machine Vision

When the train is running,the carbon slide of the pantograph is in contact with the catenary,and long-term contact will cause greater wear and tear of the carbon slide of the pantograph.Affected by extreme weather,the pantograph head may be missing and the pantograph may be invaded by foreign objects such as floating garbage.All of the above factors will cause the pantograph to fail to obtain current from the catenary properly,thus affecting the safe operation of the train.At present,the pantograph monitoring system used by railway bureau is mostly to install cameras in stations or EMU entry and exit warehouses to monitor pantograph and judge whether the pantograph is abnormal,and there are few applications to monitor and warn pantograph in train operation.In this paper,a set of abnormal state detection system of highspeed railway pantograph based on machine vision is designed to realize real-time detection and record of pantograph state during train operation,and improve the safety of train operation.Firstly,the overall structure of the detection system is determined.The camera device is installed on the roof of the train to monitor the pantograph in real time,and the image processing equipment is installed in the interior of the train to analyze and judge the data in real time,and the pantograph fault state can be alarmed and prompted through text and voice,and the alarm results can be output to the driver’s cab or the display screen in the equipment room for display.According to the technical conditions and processing scenarios of the system,the hardware selection of the image acquisition module of the system and the installation mode are determined.The Qt Designer software is used to design the working interface of the system: It mainly includes seven modules: login module,main interface,foreign body intrusion limit,wear analysis,bow Angle loss,video playback and data analysis.It uses man-machine interface to display abnormal status in real time and record abnormal data.Then the detection algorithms of the wear and bow Angle loss of the skateboard are studied.Firstly,Wiener filter is used to blur the image,secondly,U2-Net semantic segmentation is used to segment the pantograph image to remove the background interference,and then Fast Approximate Nearest Neighbor Search Library,FLANN’s feature matching technology locates the carbon skateboard region and bow Angle region.For the carbon skateboard region,improved Canny edge detection algorithm,morphological processing,edge tracking and wear calculation algorithm are used to obtain the wear curve of the carbon skateboard.For the bow Angle region,inter-frame differential detection algorithm is used to determine whether the bow Angle is missing.The adaptive threshold algorithm is used to determine the threshold T of the difference between frames.Finally,foreign body intrusion detection was carried out in the pantograph region obtained after semantic segmentation detection.YOLOv3 algorithm was used for target detection of foreign objects.For the problem of low detection accuracy of small targets,the 4th scale network prediction was added,and the K-Means ++ algorithm was used to recalculate the prior box.In order to improve the detection speed,depth-separable convolution was used to replace standard convolution,so as to build a pantograph foreign body penetration detection model,train the data set and perform experimental verification.The results show that the improved YOLOv3 target detection algorithm has accurate and fast detection capabilities,that is,it also ensures the detection capabilities of small targets,medium targets and large targets,and can quickly and effectively detect foreign objects on the pantograph.