Research And Application Of Rail Displacement Detection Based On Machine Vision Technology

In the safety detection of railway system,it is necessary to detect the rail displacement for a long time due to the existence of rail crawling and other hazards.By studying the method of rail displacement detection based on machine vision technology,a rail displacement detection system was established.The results of indoor simulation experiments verify the performance of the system.Firstly,the methods of improving target image resolution and evaluating target image quality were studied.Through the research of image resolution enhancement algorithm in interpolation,learning and reconstruction,the results show that the image super resolution reconstruction method based on self-exemplars without external training set is the best method to improve the target resolution.The experimental results show that the peak signal-to-noise ratio of the reconstructed target image quality is the highest and the target edge is easier to distinguish.Then,a method of target location based on compound location principle was proposed.The general target detection method is not accurate enough in the detection of rail creep displacement with large field of view and small target.The researches show that the method of deformable diversity similarity based on one-way nearest-neighbor matching has less complexity and significant effect.The experimental results show that the matching accuracy of this method is very high and the matching time is less than 1.5s.Next,the proposed algorithm based on circle fitting can accurately calculate the centroid and diameter of the target.Finally,an on-line detection system of rail displacement was established.The rail displacement detection model based on the optical imaging principle was analyzed.The system collected two groups of images for simulation experiment.The maximum error between the detection result of the first group of images and the actual displacement is 0.12 mm which is 18.1% of the physical size of a single pixel.So the system is feasible.The maximum error between the target centroid coordinates measured in the second set of images and the target centroid coordinate of the initial image is 0.23 mm.The maximum error is less than the physical size corresponding to 0.5 pixels.The results show the stability of the proposed target location method.The experiments verify the accuracy and stability of the system.Figure 29;Table 6;Reference 46...