| Rail fasteners are important parts used to fasten rails and sleepers,and are the basic guarantee for maintaining the normal operation of the subway system.At present,our country’s subway rail fasteners mainly rely on manual inspections,with low work efficiency and heavy reliance on workers’ experience.It is difficult to meet the urgent need for high-quality daily operation and maintenance of the rapid development of urban subways.The automatic inspection of non-contact track based on image recognition can overcome the above shortcomings.In recent years,it has attracted the attention and research of railway departments and experts and scholars.It can be seen that the research on visual inspection of track fasteners has important scientific significance and engineering application value.The thesis aims at designing and realizing an accurate,stable and efficient visual inspection system for subway track fasteners,and carries out related research.The main work is as follows.First,a three-dimensional data acquisition experimental system was built.After fully comparing and analyzing the 3D imaging technology,we chose the lidar 3D camera based on i To F technology as the core to build the hardware system,developed the corresponding supporting cross-platform software DCS-3D,and realized it by using memory cache technology and multi-threading technology to acquire three-dimensional depth images and color images of fasteners High-speed and high-quality.Then preprocess the collected data and reconstruct the color three-dimensional point cloud.Aiming at the characteristics of the collected data,an improved image preprocessing algorithm based on pyramid iterative steer kernel regression is proposed.Experiments show that this preprocessing method can remove most of the noise while maintaining image details and effectively improve the data quality.The camera calibration method is studied.Based on Zhang Zhengyou’s calibration method,the depth camera and the color camera are calibrated separately,and the depth image and the color image are registered,so as to compare the effective information of the depth image and the color image;reconstruct the color three-dimensional point cloud to realize the intuitive display of the scene.Then a "two-step positioning" fastener detection and positioning method is proposed.Aiming at the shortcomings of limited information in two-dimensional images and greater impact of light pollution,a coarse positioning algorithm for fasteners based on depth images is proposed.The improved Hough transform method is used to detect the track lines to correct the image,and the gray-scale projection method is used to locate the position of the sleeper to find out the general area of the fastener;use the improved template matching method to accurately locate the fastener in the color image,and classify missing fastener.Experiments show that this method can effectively extract fastener images and distinguish the missing fasteners.Finally,the defective fasteners are identified.Aiming at the problem of few missing fastener samples in the data,a sample library expansion method based on image symmetry is proposed,and a fastener classification sample library is established;combined with sample characteristics,a support vector machine classification method based on fusion features is proposed to identify defective fasteners.Experimental results show that this method is significantly better than the single feature classification method,can accurately identify defective fasteners,and meet the design requirements.The thesis designs and implements a three-dimensional detection system for subway rail fasteners through research on the construction of data acquisition experimental system,data preprocessing,fastener detection and positioning,and fastener defect identification.The experimental results verify that the experimental system designed in the thesis can accurately and stably detect defective fasteners,achieve the design goals,and has a good application prospect. |
PDF Full Text Request