Study On Crack Detection And Cluster Analysis Method From Straddle Monorail Track Beam Surfaces

Straddle-type monorail transportation is an important way of urban transportation.The prestressed concrete track beams forms a track of bearing train operation,the health status of track beams directly affects the safety of train operation.The surface cracks of track beams are inevitable defects in the course of operation,so the detection of surface cracks of track beams is very important work for daily maintenance.According to the datas collected from acquisition system of the research group,this paper studies the crack identification method based on geometry of surfaces and cluster analysis.In order to distinguish the image datas of finger shaped plate collected by the acquisition system,We can detect edge information by edge detection method,then distinguish this kind of data by edge information.Moreover,The image size of the track beams is very large,and the crack detection of large images can be transformed into the crack detection of many small images by using the image block and combination scheme.Firstly,the crack image can be regarded as a curved surface.On the surface,the crack is represented as a canyon.The identification of the crack image is converted to the problem of surface shape recognition by using the method of geometry of surfaces.The geodesic distance maps reflect the geometric features by calculated from four directions on the surface.The distance difference matrix is obtained from the geodesic distance maps for quantify those features,the shape descriptor be calculated according to the difference matrix,the shape descriptor is used to quantitatively describe the shape of the surface,and then to judge whether the image is a crack image.Then,using the FCM(Fuzzy C-Means)clustering algorithm clustering algorithm to segment the crack image,The traditional FCM algorithm is used to cluster the gray value of the crack image,it ignores the spatial relationship between pixels.The FCM clustering algorithm based on distance correction takes into the influence of spatial location information on clustering,which improves the robustness of image segmentation.However,there are still some noise and pseudo cracks in the results of segmentation.According to the spatial characteristics of the cracks,aspect ratio and regional vector of crack are constructed to remove most of the noises and pseudo cracks.Finally,using DBSCAN(Density-Based Spatial Clustering of Application with Noise)clustering algorithm to extract the crack after image segmentation.DBSCAN clustering algorithm clustering pixels based on density in space,it can be used to cluster clusters of arbitrary shape,and be used to cluster disconnected regions of crack.The experimental results show that the detection method of this paper provides reliable reference for the maintenance of track beams.