Rail Surface Defect Detection Based On Three-dimensional And Color Information

In recent years,the speed of trains has continued to increase,and the quality requirements for rails are also rising.Rails are the key equipment for the full weight of the train,and minor defects can cause major accidents.In order to minimize the hidden dangers in railway transportation,the rails must be inspected and the rail defects must be discovered in time.At present,the domestic surface defect detection of rails is still mainly based on manual detection methods.Therefore,this paper proposes a method based on the saliency of the rail surface defects detections using the three-dimensional depth information of rail and the color image.Study this from the following aspects:(1)Aiming at the complexity of the linear array camera imaging model and the difficulty of calibration,a simple and flexible line-scan camera calibration method is proposed.A new target pattern with orientation coded information is firstly proposed,which is composed of several sets of circular and rectangular stripes and obtain more effective coded information from various directions.The target pattern is a combination of a ring with coded information and a rectangular stripe,which can obtain different coding information from various directions.The coded information can help determine the relative position between the line camera and the calibration plate.And then the internal and external parameters of the line camera are solved based on the invariance of the cross ratio.Finally,the final calibration result is obtained by using the nonlinear optimization method according to the initial parameters.The simulation experiment and the real experiment of the line-scan camera calibration method proposed in this paper are carried out,and compared with the existing line-scan camera calibration method.(2)In this paper,the problem of rail surface defect detection is transformed into a saliency problem of rail surface detection.A saliency detection algorithm that fusion global contrast and local constrained linear coding is proposed.Firstly,multi-scale superpixel segmentation is performed on images by simple linear iterative clustering(SLIC)method.The method uses RGB features,CIE-Lab features,LBP features,HOG features,and depth information for the saliency detection to ensure the comprehensiveness of the detection and avoid the missed detection.Then,the rank-1 constraint is used to obtain the global saliency map by multi-scale saliency map fusion.Finally,the local saliency map and the global saliency map are combined to obtain the rail defect map.Multiple integrations can improve the accuracy of the detection algorithm and reduce false detections.In the laboratory environment,the saliency detection algorithm proposed in this paper is experimentally verified.The saliency images at each stage are extracted for comparison.Finally,the experimental comparison with the existing saliency detection algorithm is carried out.(3)This paper designs and implements the rail surface defect detection system.The 3DPIXA color line-scan camera is used to capture the surface image of the rail.The Corona II light source solves the lighting conditions on the surface of the rail.We can adjust camera exposure,gain and light source brightness according to the needs of the site environment.The encoder is used to achieve the matching speed of the moving platform and the line-scan camera,and a uniform and stable high quality image is obtained.Finally,the obtained rail surface image and the corresponding depth map are used to detect the defect using the salient detection algorithm proposed in this paper.