Track Image Recognition Based On Improved Canny Operator

As a national strategic infrastructure,transportation plays a role in promoting economic growth.Railway is the main artery of national economy and the key to invigorate the economic development of various regions.Therefore,in order to ensure the smooth operation of the train and improve its safety and reliability,it is imperative to add foreign body intrusion detection and protection system in the train control system.In the foreign body intrusion detection system,one of the important steps is to reasonably define the intrusion detection area,that is,to take the rail as the reference point,to form the outward extension area and the inner closed area.The purpose of this paper is to accurately delimit the intrusion detection area at the same time,which can not only reduce the time cost of the whole algorithm,but also ensure the effectiveness of the detection algorithm to a certain extent.However,in the current research,foreign scholars mostly use sensor technology to realize disaster prevention and mitigation monitoring of railway track,but few use image processing technology.In addition,most of the track detection algorithms proposed by domestic scholars are oriented towards the structured straight or curved track,ignoring the situation of the turnout on the railway.In view of the shortcomings of the current research,this paper proposes to use the image processing technology to realize the railway track detection.The edge distribution of the track is detected by the edge detection operator,and the Therefore,there is a big difference between the edge of the track and the edge of the surrounding environment.Firstly,the image pre-processing technology is used to reduce the noise of the original image.In the process of track edge detection,the track distribution is not simple horizontal or vertical,and it is easy to lose edge pixels by simply calculating the pixel gradient values in the horizontal and vertical directions.In this paper,gradient templates in both directions of 45° and 135° were added on the basis of traditional Canny operator to ensure that more true edges were detected.In this paper,three algorithms are used to detect the most common straight rail,curved rail and single turnout.The implementation of the three algorithms is different.The difficulty of straight track detection is to distinguish the left and right tracks,which is mainly based on the positive and negative slope of the straight line.The difficulty of curved track detection is also to distinguish the left and right tracks,but the slope distribution of the curved track is different from that of the straight track,which is mainly based on the two characteristics of the slope distribution of the straight line and the distance between the straight lines.The difficulty of fork detection is to distinguish the straight line and curved line.In the algorithm,the long line in the edge image is obtained as the straight line edge by adjusting the threshold value of Hough transform,and then the short line in the edge image is obtained by adjusting the threshold value of Hough transform.The short line in these short lines that conform to the distribution characteristics of long line is the curved line edge.The experimental results show that the improved Canny operator proposed in this paper can detect the edge of the track more abundantly.The detection algorithm for straight track,curved track and single open switch can also accurately identify and fit the track.