Rail Corrugation Measurement Method Base On Binocular Vision And Its Application

With the rapid development of computer vision,computer vision is widely used in real life and industrial production.Binocular stereo vision is an important research field of computer vision.It has the advantages of simple structure,non-contact and low cost by simulating human eyes for scene reconstruction and measurement.It can avoid the problems of surface wear and deformation caused by contact with measured objects,and has more accurate measurement accuracy than estimation by human eyes.Binocular vision technology has broad application prospects in unmanned driving,virtual reality,precision measurement,three-dimensional reconstruction and other fields.In this paper,the key technologies of binocular stereo vision are studied,and an automatic segmentation method of binocular images is proposed to reduce the computational complexity of stereo matching algorithm and improve its efficiency.At the same time,an adaptive binocular stereo matching method is proposed for objects with different colors and textures.This method can accurately obtain disparity maps and three-dimensional point clouds.The results show that the method designed in this paper can achieve high measurement accuracy in processing and analyzing three-dimensional point clouds.Aiming at the problems of inefficiency,easily influenced by human factors and inconvenient storage of results in traditional rail corrugation measurement,the application of binocular stereo vision technology in rail corrugation measurement is studied.A rail corrugation measurement system based on binocular stereo vision is designed and each functional module is realized.The main contents of this paper are as follows:(1)Binocular camera calibration: This paper adjusts the camera angle and illumination environment of the circular array calibration board to the best state,and uses OpenCV visual library for automatic calibration.In the calibration process,the calibration parameters whose RMS value of re-projection error is less than 0.2 pixels are selected as the calibration target.(2)Binocular epipolar correction: This paper uses C++ language to realize epipolar correction based on Bouguet algorithm,so as to reduce the computational complexity and difficulty of subsequent stereo matching algorithm,and SURF feature detection algorithm is used to verify the accuracy of Bouguet epipolar correction.(3)Effective region extraction: In order to reduce the computational complexity of stereo matching algorithm and improve the efficiency of corrugation measurement,this paper uses a combination of point laser and GrabCut segmentation algorithm to automatically segment interested image regions.The sub-pixel coordinates of the laser center of mass are extracted by morphological method to determine the rectangular frame position required by GrabCut algorithm and the disparity value needed to be increased when the image is saved as a new image after segmentation,so as to ensure the disparity invariance.Experiments show that the method effectively improves the efficiency of stereo matching.(4)Adaptive binocular matching algorithm: This paper uses convolutional neural network to train Middlebury standard data set,and obtains the cost network to calculate the matching cost value while using fast guided filtering to aggregate the cost as the data item of the alpha expansion algorithm.The tilted window model is used to overcome the disparity forward parallelism problem.Each pixel is regarded as a point on a plane in three-dimensional space,and the plane parameter is a 3D tag.This paper combines space propagation,view propagation,plane simplification and alpha expansion algorithm to solve the best 3D label for each pixel position,and proposes the idea of image block grouping combined with OpenMP parallel technology to speed up the algorithm.The experimental analysis show that this method can obtain accurate disparity map and three-dimensional point cloud for Middlebury standard data set,rail image and various objects collected,thus has good adaptability.(5)Experiments validation: This paper designs and implements each module of rail corrugation measurement system based on binocular vision.The accuracy of the algorithm is verified by using standard gauge block and vernier caliper.The error can be effectively controlled within 0.1 mm,which meets the accuracy requirements of the Ministry of Railways for rail corrugation detection.At the end of the paper,the corrugation of the test rail is measured and the curve of the rail is drawn.This paper has the following innovations:(1)An automatic rail image segmentation method based on GrabCut is proposed.This method can segment the interesting part of the rail image and save it as a new image.It can reduce the computational complexity and guarantee good quality of disparity map.(2)A binocular adaptive matching method based on 3D tag computation is proposed.The idea of combining alpha expansion algorithm with view propagation,space propagation and plane simplification is proposed to calculate accurate disparity map.The idea of image block grouping and OpenMP technology are combined to improve the matching speed through parallel computing.This method has a good disparity map effect for objects with different color,shape and texture information,and meets the adaptability of the matching algorithm.