The Research On Detection And Tracking Algorithm Of Pantograph And Contact-wire Based On Monocular Infrared Images

The higher requirement for intelligently monitoring the working state of the contact-wires and the pantographs of the moving trains is already presented with continuous railway speed-up in China,especially in very high speed state.A new technology of non-contact intelligent detection and tracking is proposed to support the rapid development of the high speed railway systems.As shown,the position character between the pantograph and the contact-wire can be detected efficiently by infrared imaging detection,which has been widely used in recent years.However,in the real catenary-pantograph systems,there are still many problems about the real-time detecting and tracking of contact-points according to infrared imaging under the high-speed condition.The main reasons include:(1)The zigzag contact-wire should be reinforced and folded by steady-arm so as to result in a relatively huge contact-point detection error at certain intervals;(2)The contact-points cannot be distinguished easily when the contact-wires are transformed frequently at typical railway turnouts.(3)The complexity of the scene makes the contact-point detection harder.Aimed at the problems of short line and line-changing during contact-point tracking period,the detecting and tracking algorithm of contact-wires and contact-points,based on the infrared images captured by the Catenary-Checking On-line Monitor System,have been deeply studied,including the following parts:Firstly,a pantograph model is constructed and a rapid locating of the pantograph is achieved by template matching.According to the brightness and width characters of the contact-wires in infrared images,an enhancement method of the contact-wires and pantograph(ECP)is proposed by using the left-right gradient and the gray value of contact-wires,which can be easily used to extract the points set of contact-wires.The line parameters are estimated by Random Sample And Consensus Algorithm,so as to effectively solve the problem of determining the contact-wire parameters,which are already folded by steady-arms,inaccurately.Secondly,based on enhanced images,a single-frame detection(SFD)and a multi-frames tracking(MFT)for detecting contact-wires are proposed,respectively.In the initialization phase,a continuous SFD with several consecutive images are used to locate the contact-wire by statistical voting and further construct the MFT model.Through checking the difference between the SFD results and the tracking model during tracking and then updating the tracking model and further combining the positional constraint of the carbon slide,the contact-wires can be switched automatically.Finally,the algorithm proposed in this thesis is tested using two sequence of infrared images.The experimental results show that the proposed algorithm can effectively utilized to deal with the short line of the contact-wires and also select the right one autonomously.Typical performances include:(1)the recognition rate of the contact-wire is 98.23% and the average pixel error 0.4 in simple clean background sequence;(2)the recognition rate of the contact-wires is 94.75% and the average pixel error 0.51 in complex background sequence and the average FPS 80;(3)the requirements of technical specifications and real-time detection of the high-speed trains can be satisfied,so as to lay a foundation for practical application of the methods proposed.