Key Techniques Of Registration And Fusion Of Rail 3D Shape

As one of the most important part of China’s public infrastructure,rail transit online measurement of rail three-dimensional shape can accomplish the purpose of real-time monitoring of rails and ensuring the safety of railway transportation.The optical three-dimensional measurement has the advantages of no contact with the measured object,measuring and operating quickly.Thus online phase measurement profilometry can be used for obtaining the three-dimensional shape of the rail.However,the range of the three-dimensional image field of the acquired rail profile is limited,and the image registration technology can be implied to broaden the field of view of the three-dimensional image of the rail profile.When the traditional iterative closest point method is applied to registering 3D images of rail,the three-dimensional shape of the rail has no obvious feature points leading to mis-registration,and the algorithm takes a long time to run.In addition,the three-dimensional images of rail have the problem of overlapping area and data redundancy after image fusion processing.In order to solve the problems in the three-dimensional images registering process of the rail,this thesis researches an improved registration method: Find the corresponding relation between the point cloud data and the three-dimensional images of the online movement rail,and then use it as the basis for rough registration of three-dimensional rail images.Next,an improved iterative corresponding point algorithm is taken advantage of accurately registration the three-dimensional point cloud data of the rail profile.This improved method consumes less time,has high precision and fits well with the registration process of the point cloud data of the rail three-dimensional shape.For the giant amount of the registered point cloud data of 3D rail shape and the exiting the overlapping area after registration process,Octree was made use of fusing and compressing the redundant data of 3D rail shape in this thesis.Through the data fusion operation,the amount of 3D point cloud data after compression is reduced to about one-tenth of the amount of data after fusion,which greatly reduces redundant point cloud data.This thesis proposes a new point cloud data fusion evaluation method: The points in the overlapping area before and after the point cloud fusion are searched for the 8 nearest neighbors by using the kd tree.And the principal component analysis is used for calculating the normal vectors of these points.Then the angle between the normal vectors before and after the fusion is calculated,which can be utilized for evaluating the fusion result.