Research On Detection Of Railway Turnout Based On Machine Vision

With acceleration of national railways and construction of high-speed railway,railway transportation has entered a new stage of development of our country, whichindicates that the cause of China’s transportation planning toward a more efficientdirection, there will be increasing requirements for track security. As a result oftoday’s science and technology development, intelligent transportation will play animportant role in the safety aspects of the construction of the railway. The researchof China’s intelligent transportation system began in the late80s in last century, inrecent years, remarkable results have been achieved, and then brought hugeeconomic and social benefits to the community. Railway turnout detection as animportant part of intelligent transportation realizing high-speed, real-time, accuratemeasurements when changing the pitch of the turnout ensure smooth traffic safer,which has far-reaching implications for intelligent railway construction.Turnout is an important part when railways operate. It is a device to transfer atrain from a group of tracks to another. Whether the turnout conversation reach thestandard will affect the safety of railway operations so that the monitoring anddetection of the turnout are referred to a new height. Our intelligent transportation isstill in its infancy, the detection of railway turnout mostly done by hand, the researchand development of related technologies become more pressing. This researchsubject is the tight measurement of turnout during the conversation. It proposes adetection method based on machine vision, which combines the image acquisitionsystem with the image processing system to process and analysis on turnout image.This research achieved the detection techniques of the turnout through experiments,and made innovations of turnout feature extraction method.Turnout feature extraction process is divided into for parts, image preprocessing,edge detection, turnout edge extraction, edge line fitting. In the part of imagepreprocessing, the use of gray-scale transformation and gray-scale enhancement ofthe images camera captured, make the edge feature even more prominent. In the edge extraction processing, this paper use an edge detection method that the programautomatically calculates the threshold based on Canny edge detection operator, inwhich high and low thresholds obtained automatically calculated by the program. Itcan reduce the simplification of thresholds set manually, and enhance theadaptability of edge detection. The result of edge detection shows that the turnoutpitch is the distance between two straight edges, so the Hough transform ideas willbe used, and it must be improved on the basis of the tradition Hough transform. First,the image pre-processing to improve the distribution of pixels, with the idea ofmatching region for pixel filter can reduce the number of pixels to participate Houghtransform. Then, using a peak reverse cleared method to ensure the uniqueness of theobject, effectively reduce the interference pseudo-edge pixel, and achieved goodresults. Because of the accuracy requirements of turnout measurement, the result ofHough transform linear fitting accuracy is not high, the area is difficult to control.Finally, it must be combined with the method of least square. The two edges detectedby the Hough transform and the edge image detected by Canny edge detectionrespectively, the result of points set as fitting line, then using the least square method.The experiment shows that the fitting line coincides with the turnout edge, and thevalues are more accurate.The accurate extraction of turnouts’ information laid a good foundation for nextmeasurements. In measuring processing, selecting some basic turnouts that are notclosed because the basic track width is a certain value, it can be as a reference. Afterusing the same detection method on the basic track, proportional relationshipbetween pixel width and reality width can be get. Thus, when the turnout closed, thedistance between two edges can be obtained and the turnout detection purpose can beachieved.Finally, this paper summarizes the word and proposes the shortcomings in theresearch, then put forward the work plan for next step.