Research On The Key Techniques Of Catenary Measurement Based On Machine Learning

Due to its visual characteristics,images are widely used in the communication of people's information.With the development of computer image technology,people are not only satisfied with the information displayed by the two-dimensional image,but also hope that the image can show a more intuitive and vivid three-dimensional sense.In general,we need to achieve three-dimensional vision technology with more than two cameras.Stereoscopic technology can greatly enhance the sense of reality,so it is widely used in production and daily life.In this paper,we make use of stereo vision technology to measure the catenary.Catenary is a special transmission grid that supplies electric locomotives with electricity.People measure the catenary to ensure the safe operation of electric locomotives.The current contact measurement methods have many deficiencies,the maintenance process is cumbersome,the operation cost is large,and the traffic safety will be affected to a certain extent.Therefore,we use the stereo vision technology to realize the non-contact detection,and detect the position information of the catenary by the line cameras.The parameters of the inspection include the height value and the pull-out value.This article proposes a new measurement method.In detail,the specific method is to install four linear cameras on the top of the electric locomotive,and obtain the image of the contact lines by continuously exposing the contact lines.Locate the pixel coordinates in the images,and then we get the predicted real space position of the contact line which is predicted by the trained support vector regression models.Finally,we evaluate this method by using annotation dataset,and improve the detection effect through fusion measurement method.The practical application of industry proves that the proposed detection method not only satisfies the detection accuracy but also gives full play to the advantages of non-contact measurement,effectively reducing the operation cost and maintenance cycle.The main work of this paper is as follows:Firstly,according to the characteristics of the images taken by the line camera,we stitch the linear images together to form a two-dimensional image that can be used for the detection,then we propose the tracking detection algorithm of the contact lines of in images.By using the image preprocessing method and the video tracking technology,we preferably exclude the noise interference and achieve the contact line detection in the complex background of the noise.Secondly,this paper summarizes the past camera calibration process and proposes a new camera calibration method that uses support vector regression to model the relationship from pixel coordinates to world coordinates.The advantage of the process is that the calibration becomes a "black box" which does not require complex parameters to characterize the model.The proposed method has the advantages of fast calibration and calibration accuracy to meet industrial requirements.Thirdly,aiming at solving some problems in wires detection,a fusion measurement method is proposed.The detection results of the wire in the image are fed back by measuring the position of the contact wire.The scheme can effectively improve the detection effect of the contact wire and improve the wire detection accuracy.