Non-contact Rail-wear Inspecting System Based on Image Understanding

Inspecting rail-wear status is critical to guarantee high-speed running safety, proper maintenance schedule, and low operating costs. A dynamic inspecting system based on machine vision is presented using near-infrared laser line projector, CCD (charged coupled device) camera and image processing technology for detection of rail wear parameters. The design of single-element laser-beam projector is described. The projector takes a Gaussian laser beam and expands it into a uniform intensity line. A calibration frame has been constructed to obtain system coefficients before detection. A DLT (direct linear transformation) algorithm is analyzed and modified to correct the nonlinear lens distortion using a planar spot-grid. To analyze the shape of rail header, the corners (key points) on rail curve, using CSS corner algorithm, have been detected successfully. Compared with relevant points on standard rail curve, rail wear values are computed fast and accurately. And hundreds of rail profile images have been acquisitioned and processed. At the same time, rail wear values have been obtained fast and accurately. It is testified that the whole system of rail wear reflects the status of rail profiles wear authentically and dependably. Having high performance /cost ratio, the system is much suitable for Chinese context.