Application of machine vision technology in quantitative three-dimensional reconstruction of rail corrugation

Aiming to achieve the quantitative reconstruction of rail corrugation, the technical route of ‘self-developed equipment - image acquisition - three-dimensional reconstruction’ based on machine vision technology is established in the present paper. Firstly, the surface and profile images of rail corrugation are captured through the self-developed acquisition equipment. Then, the image processing techniques are used to extract grayscale values and wear depth information from the captured rail surface and profile images. Subsequently, the fitting model correlating grayscale values and wear depth of rail images is developed by neural networks, and the three-dimensional reconstruction and validation of rail corrugation are carried out through field tests. Results indicate that the correlation coefficients of fitting model are 0.99146, 0.99149, and 0.98798, with overall errors are primarily distributed around zero. The average error of the rail corrugation quantitative reconstruction model is 0.003 mm, which demonstrates the accuracy of the three-dimensional reconstruction method for rail corrugation based on machine vision. This research offers a theoretical foundation and technical support for quantitatively characterizing rail damage.