Research On Track Inspection System Based On Machine Vision

The track gauge transfinite would cause train running safety problems, which was animportant parameter in evaluation of the track quality. Therefore the track gauge parametersneed to be periodic testing and the track need to be timely maintenance. In order to solve theinefficiency, poor precision and invalid data recorded phenomenon of artificial gaugeinspection on existing line, make up the shortage of the domestic inspection capabilities intrack inspection vehicle. Rail cars could be used for the inspection platform in mechanic ward,and a schema style track inspection system based on machine vision was established, a depthstudy of the inspection of track gauge parameter had been finished.In view of the algorithms of image processing and vibration correction were difficult todraw from the domestic introduction of the visual track inspection system. A complete imageprocessing algorithm constituted by gray-scale transformation, filtering, image segmentation,morphological transformation and centerline extraction in extraction of the single-pixel trackcontour was presented by this thesis, the image morphological transformation algorithm wasoptimized by run length encoding.The calibration of the visual track inspection system had been studied. The result of thegauge inspection was analyzed which was influenced by vehicle body oscillation. Proposedthe vehicle vibration component of the traveling direction, horizontal and vertical had noimpact on the gauge inspection accuracy, correction formula was deduced to correct theinfluence of the measurement results by the angle posture change in roll, pitch and yaw.A laboratory verification system had been constructed in the software and hardwareaspects based on the overall system design. Traditional schema style gauge inspection systemwas developed by on-chip and text compiled which led to professional dislocationphenomenon. Vision and LabVIEW were proposed to build image processing model andsystem software architecture which were more consistent with engineering application, thenthe research and development cost had been reduced and the development cycle been shorten.The needs of engineering and technical personnel at the scene operation could be met byman-machine interface and data storage and analysis module.The system performance evaluation experiment had been done by the laboratoryvalidation system. From the experimental results: The processing speed of the track contoursingle frame image had been increased for5times by the run length encoding optimizationalgorithm, the system measurement accuracy was±0.8mm which met the testing standardsrequirements. The mean square error of the gauge measurement results after vehicle bodyoscillations correction was0.962mm, the system maximum combined uncertainty underlaboratory conditions was0.52.