Edge Extraction And Geometry Calculation Of Railway Track In Close-Range Images

The railway track is the basis of train operation and plays an irreplaceable role in railway transportation. In the process of train operation, tracks directly bear the load from the wheels, and then pass it accordingly to subgrades, bridges or tunnels. In order to ensure a safe and stable train operation according to the designed speed, strengthening the inspection and maintenance of the railway track is necessary.The static geometry inspection of railway track is usually performed by field surveying point by point with the track geometry measuring instrument and the total-station. It has high accuracy but high labor intensity at the same time. This thesis introduces the close-range photogrammetry into track detection and conduct track geometry inspection with the computer vision and digital image processing.Since the gray difference between the area of track and the surrounding region in digital close-range images is not obvious, this thesis uses the logarithmic transformation for image enhancement, the Canny algorithm to extract the edge of track, and then uses the probabilistic Hough transformation and least square method to fit the edge line of track. The experiments show that the mean fitting precision can reach 0.98 pixels and the fitted edge agrees well with the actual edge of track.Considering the close-range track images have the certain characteristics such as gray similarity, this thesis uses the ORB algorithm to extract feature points of track surfaces between adjacent images, and uses KNN algorithm to match homonymy points, and then establishes a polynomial model with the homonymy points. The experiments show that in this model the mean precision of check points can reach 1.12 pixels in x coordinates and 0.80 pixels in y coordinates. This model can be used to well express the relationship between track surfaces in the adjacent images.This thesis establishes a spatial model of track edge line by using the mathematical relationship of track surfaces and the points in track edge lines, and this model can be used to obtain the homonymy points in track edge lines. Then the 3D coordinates of edge points can be calculated by the forward intersection which uses the results of self-calibration bundle block adjustment processing. The experiments show that the matched homonymy points in edge line has high accurate and can be used to calculate the spatial coordinates.This thesis calculates the railway center line by using the 3D coordinates of edge points of railway track, and then projects the center line to different planes to obtain the planar and vertical profiles of railway line. The experiments show that this method can accurately calculate the 3D coordinates of center line and draw the planar and the vertical profiles of railway.