| Railway is an important infrastructure related to national economy and people’s livelihood,with the characteristics of high speed,convenience and comfort.The settlement of railway subgrade is very important to the safe operation of railways.With the rapid development of railways,traditional geodetic surveying methods and civil engineering monitoring technologies cannot meet the needs of railway operation and management.Machine vision has the characteristics of non-contact,real-time,good visualization and intelligence.It has a wide range of application prospects in engineering practice.In this paper,based on the application of machine vision in the railway subgrade settlement monitoring system,the key technologies in image-based subgrade monitoring system are studied.Firstly,the traditional subgrade settlement monitoring methods and the application of machine vision in related fields is analyzed.Basic concepts of spatial geometric transformation and the relationship between image coordinates,camera coordinates,world coordinates and camera models are introduced.Position-pose measure based on machine vision is the basis of the position and attitude transmission in the camera network,so the technology of position measurement based on machine vision is mainly introduced.Combining the characteristics of laser collimation and the posture measurement technology based on monocular vision,the principle and structure of the image-based railway subgrade settlement transfer camera network are introduced.Secondly,aiming at the problem of position-pose calibration between cameras without overlapping field of view in the settlement transfer camera network,the pose measurement model is established.Each camera in the monitoring station must face the corresponding monitoring target surface,so there is no public field of view between the cameras,which brings difficulties to the calibration of the pose relationship between cameras.Four feature points with square distribution were set on the monitoring target surface.Camera rig is moved twice in small step and take the pictures of the target surface in three different position.The camera movement trajectory was obtained by solving the position-pose relationship between the camera and the target surface with P4 P algorithm.The relationship between the camera and the target surface is solved by P4 P algorithm,and then the camera’s moving trajectory is obtained.Based on the properties of fixed positions between cameras,the pose equation is constructed.The position-pose matrix between cameras was obtained by matrix rearrangement method,and the result is nonlinearly optimized by levenberg-marquardt algorithm.The simulation experiment results indicate that when the noise variance is less than1 pixel,the angle error is less than 0.02° and the translation error is less than 0.06 cm.Finally,as the camera parameters are affected by temperature,the accuracy of settlement monitoring will be reduced.Combined with camera pinhole imaging model,we deduced the relationship between image drift and change of camera’s parameters.By extracting the center coordinates of mark point on calibration plate at different temperatures,the change of the camera parameters at different temperatures is obtained.Taking temperature as input and the changes of camera parameters as output,a camera temperature compensation model based on PSO-RBF neural network is established.The accuracy of the model is verified through experiments.Pose parameters between the camera and the calibration plate at different temperatures is obtained through the corrected parameters,which verifies validity of the model. |
