Research On Railway Scene Recognition And Fastener Location Based On Computer Vision

Since China officially entered the era of high-speed railways,the safety of railway operations has always been an important issue that has an impact on the country’s economic development,safety,and national economy and the people’s livelihood.In order to ensure the safety of the railway,the detection of a series of railway infrastructure is particularly important,including the detection of rail defects and fastener defect detection.The location of the rails and fasteners mainly depends on the prior knowledge of geometry:the tracks are equidistant in the middle part of the image and the fasteners are close to both sides of the tracks,and then the template matching is performed in the predicted range.However,there are a lot of turnouts and ballast scene in railway lines,which can not make use of the geometrical a priori of the track lines.Therefore,this paper will focus on the task of identifying the turnout scene and the ballast scene.For the location of railway fasteners,the previous template matching method of the research group used geometric priori to improve the efficiency of sliding window search.However,in the scene of turnout or ballast scene,the geometric priori can not be fully utilized and can only be searched globally,which can not meet the needs of real-time detection.Therefore,we still need a better method for locating fasteners in turnout or ballooning scene.With the success of deep learning in the field of computer vision,this paper attempts to use deep learning techniques to solve complex scene fastener location issues,hoping to improving efficiency and accuracy.The main work of the paper:1.A turnout recognition method based on the geometric characteristics of turnout is proposed.This method uses the feature that the turnout image contains many tracks,while the non-turnout image has only one track.A method based on geometric features is proposed to identify the turnout scene.In this method,the line detection algorithm of LSD(Line Segment Detector)and the Hough line detection algorithm are used to detect the rail,and the discriminant criterion is given to the detected line to determine whether it is a turnout or not.Experimental results verify the effectiveness of the proposed method.2.The method of ballast scene recognition based on texture features is proposed.The method takes advantage of the clutter of the stones in the scene,extracts the edge features and performs statistical analysis to determine whether it is a ballast scene in the image.The experimental results verify the effectiveness of the proposed method.3.Fastener location method based on depth learning is proposed.Firstly,all the fasteners in 1000 tracks are collected and marked as training set and trained on YOLOv2 network.Then,the test image is identified as turnout scene.If it has one track in the images,the rail and the two sides of the fastener possible position range as a candidate area,otherwise,the whole map as a fastener candidate area;Finally,the candidate area using depth learning network for fastener positioning.The method has achieved good experimental results and realized the fastener positioning in the turnout area.