Research On An In-situ Stereo-vision System For Railway Turnout Inspection

In the process of high-speed railway construction and speed increase of railway trunk lines,railway safety inspection is an important means to ensure the normal and safe operation of trains.The turnout is the conversion equipment for the change of the railway line.as an important part of the high-speed railway,it is also the weakest link in the railway track structure.the safety of the turnout directly affects the safety and speed of the train.its reliability is one of the important factors affecting railway safety,especially the tightness of the turnout rail directly affects the traffic safety,so it is particularly important to accurately detect the turnout parameters.The traditional way of regular manual inspection of turnouts is greatly affected by artificial subjectivity,the measurement results are not accurate,easy to cause false detection,and can not carry out real-time on-line detection.The use of machine vision three-dimensional measurement technology can quickly and accurately carry out real-time on-line detection of turnout parameters,improve detection efficiency and accuracy,and meet the requirements of intelligent turnout testing equipment.In this paper,a set of on-line detection system is designed for the three parameters of the turnout tip rail: density,repulsion and creeping,and the binocular vision three-dimensional measurement scheme is adopted to overcome the influence of external light by actively projecting near-infrared laser to the turnout surface.the embedded board is selected as the system image processing module to calculate the turnout parameters in real time and return the calculation results to the host computer in time through network communication.The main contents of this paper are as follows:First of all,in view of the complex environment of the railway turnout site,the requirements of the actual turnout inspection process are analyzed in detail,and the binocular vision method with the characteristics of simple structure,no mechanical wear and high precision is adopted as the system imaging scheme.at the same time,the line laser is used to actively throw ray laser to the surface of the turnout to increase the characteristics,which solves the problem that the surface of the turnout is smooth and the feature is not obvious.It is easy to cause the failure of feature matching in the process of binocular vision stereo matching.The near infrared band laser is selected in the laser selection,which avoids the influence of the change of external illumination on the imaging quality.Then,the camera imaging model and the single-double target determination method are studied.by calibrating the internal and external parameters of the camera and correcting the lens distortion,the external parameters of the binocular vision system composed of the left and right cameras are calibrated,and the left and right camera coordinate systems are unified.determine the conversion relationship between the camera coordinate system,the image coordinate system and the world coordinate system.In the aspect of light stripe center extraction,an improved directional template light strip center extraction algorithm is proposed on the basis of grayscale center of gravity method and directional template method: firstly,the light stripe region is obtained by image preprocessing filtering and edge extraction.then the light stripe skeleton is refined and extracted,and four directional templates in different directions are designed for sliding convolution of the light stripe skeleton.In the direction of the largest convolution response in the four direction templates,the gray center of gravity of the light stripes is calculated for each column as the central coordinates of the light stripes,which solves the problems of slow speed and large deviation of the results of the traditional light stripe center extraction algorithm.The binocular stereo matching algorithm is studied.In order to meet the real-time requirements of turnout detection,the combination of epipolar constraint and sequence constraint is used to match the center point of light stripes in the left and right images.Through the triangulation method combined with the system calibration results,the three-dimensional coordinates of the center point of the light stripe are calculated,a reasonable geometric model is established,the data of the center point is fitted,and the density,repulsion degree and creeping quantity of the turnout are calculated.Finally,the functional requirements of the system software are analyzed,and the software interface of the binocular vision turnout detection system is developed under the Windows operating system,and the software is tested.A large number of experiments are carried out for different states of turnouts,and the measurement accuracy can reach 0.5mm.When the density,repulsion or creeping capacity of the turnout changes,the system can update the measurement results immediately.The experimental results show that the system designed in this paper can meet the actual requirements of railway turnout detection.