Video-based Pantograph Arc Detection In Urban Rail Transit

With the rapid growth of the domestic economy,urban rail transit is widely popularized in major cities.The operation scale of urban rail transit continues to grow,and the requirements for safe and stable operation are getting higher and higher.The relationship between pantograph and catenary is an important part of train operation.The occurrence of pantograph arc will accelerate the loss of pantograph,affect the power supply of the train,and generate electromagnetic interference.Therefore,the monitoring of pantograph arc is of great significance to the safe operation of trains.This paper studies the existing pantograph arc detection technology at home and abroad,and there are common problems such as complex detection equipment,high cost and low detection efficiency.This paper only relies on the pantograph-catenary surveillance video taken with the train as the sample material,and proposes a pantograph-catenary arc detection model based on correlation filtering and neural network.The main research contents of the paper are as follows:Firstly,sample analysis and image preprocessing are performed on the pantograph-catenary surveillance video collected by the train,and the prominent features of pantograph-catenary arc images are analyzed through image preprocessing,and preliminary detection is carried out based on the characteristics of pantograph-catenary arc images in the video time series.Before detecting the pantograph arc,in order to eliminate the interference and reduce the amount of calculation,the tracking algorithm should first be used to intercept the part of the pantograph contact point where the arc is generated.According to the architecture,the correlation filters are trained by different features,and the gradient feature filter that can stably track the contact point of pantograph and catenary and the color feature filter that is sensitive to pantograph arc are obtained.At the same time,the tracking and detection strategies of the two filters are designed.The experimental test shows that the pantograph-catenary arc tracking detection model based on correlation filtering can stably track the pantograph-catenary contact point and accurately detect the pantograph-catenary arc.In order to improve the recognition accuracy of pantograph arc and realize the classification and recognition of pantograph arc,this paper designs a pantograph arc detection and classification model based on neural network.Due to the uncertainty and small probability of arc occurrence,it is difficult to collect a large number of pantograph-catenary arc pictures on the same line as the training set of the neural network.By introducing a generative adversarial network to expand the data set,and using correlation filters to constrain the generated pictures,The availability of generated images is guaranteed.Based on the structure of residual neural network,a neural network model for pantograph arc recognition is built,and the input data processing part and residual part of the network are improved to accelerate the convergence speed of the neural network loss function.In the experiment,the accuracy,detection speed and structure of the correlation filtering and neural network models are compared.The results show that the two models have better detection performance,and the advantages of the two models are analyzed in various aspects.In practical applications,flexible choices can be made according to demand scenarios and application environments.