Binocular Stereo Vision Based On Railway Automatic Hook Removal Research On 3D Reconstruction Technology

With the rapid increase of railway freight volume in recent years,railway automation technology has been put on the agenda.Now,all kinds of large train marshalling stations,such as Zhengzhou North Station and Nanjing East Station,have basically realized railway transportation automation,which greatly improves the efficiency of train shunting and marshalling.However,the decoupling of the train at the hump still needs to be carried out manually by the staff,and there are many disadvantages in the manual decoupling operation,such as:it is easy to take off the hook in the fast decoupling time,slow down the marshalling process,affect the efficiency,and there are great personal safety risks.In order to realize the automatic pick-up,this paper designs an automatic pick-up robot,implants "brain" for the robot,collects the binocular image of the freight train at the hump through the binocular stereo vision system,and uses the binocular parallax principle to measure the physical distance between the vision system and the handle,coupler and carriage,finally obtains the three-dimensional information of the scene environment,and completes the three-dimensional reconstruction.According to the optical principle of the camera,the imaging model of the camera is established.In order to solve the model parameters,Zhang Zhengyou camera is used to calibrate the camera for secondary calibration,and the accurate camera parameters are finally obtained.The goal of this study is to recover 3D information from 2D images,so we need to recover the pose information of binocular camera.In this paper,SURF,SIFT and ORB feature points of binocular image are matched by improved optimized nearest neighbor matching algorithm.According to the polar geometry principle of binocular camera,the position relationship between binocular cameras is established and the pose information is recovered by using the calibrated camera parameters.Because the train hook removal work is carried out outdoors,the external environment has a great impact on the image quality of the camera,so it is necessary to preprocess the image.On the basis of traditional image preprocessing,this paper proposes an adaptive sharpening fusion algorithm based on the relationship between the 8 neighborhood of image edge pixels to enhance the quality of the image and increase the number of feature points that can be detected while removing the noise.Finally,on the basis of enhancing the image quality,the target recognition is realized by template matching.Stereo matching to obtain the scene disparity map is the focus of this study.Firstly,through stereo correction,the binocular image pixel registration is realized.Secondly,the weight of SSD cost function and Census cost function is combined to reduce the influence of noise and uneven illumination in stereo matching,and the parallax map is filtered by means,so that the texture of disparity map is clear and the segmentation area is obvious.Finally,using the optical center distance between cameras(baseline)and parallax,the depth Z is obtained by triangulation,and the point cloud image of the scene is displayed through PCL point cloud library.