Detection Of Rail Surface Defects Based On 2D And 3D Visual Information

The development of China's railway is in the golden time.It requires a large number of rails,so the accuracy and speed of rail quality inspection need higher requirements.At present,there are number of methods that detect rail surface quality only based on two-dimensional image features or based on three-dimensional point cloud alone.However,defect detection based on two-dimensional image features are disturbed by oil stains and oxidatio n easily.Some defects which are not obvious are also difficult to identify.3D point cloud registration can make up for the Shortcomings of two-dimensional inspection,but can not detect crack defects.This paper studies the rail surface quality inspectio n technology based on 2D image features and 3D point cloud fusion.According to machine vision principle and factory rail surface quality inspection standards,this paper research a set of rail surface quality inspection equipment,which use the technology of two-dimensional and three-dimensional visual information combination,it contains three-dimensional reconstruction of rail surface contour,defect segmentation and defect classification.The geometrical model of the three-dimensional reconstruction of rail surface with grating coded optical sensor is established.The calibration of the measuring system is completed by Zhang Zhengyou calibration method and the conventional binary grey code and the invariance of cross ratio.The internal parameter of the camera and the grating plane parameter are obtained.Encoder completes the combination of two-dimensional information and three-dimensional information.Using BP neural network to identify the four kinds of defects.For crack defects which the method of three-dimensional can not be identified,it can use crack characteristics such as area,length,circumference,and determine whether the crack is or not ultimately.The experimental system of rail inspection combined with 2D and 3D visual information was carried out and the algorithm of this paper was tested systematically.The results show that defect recognition technology proposed in this paper can identify the rolling scar,rolling ma rks,Inclusion,dent,crack defects accurately.