Design And Implementation Of Track Plate Geometry Detection System Based On 3D Reconstruction

The traditional track plate measurement system uses steel rulers,vernier calipers,and total station to complete the measurement.It takes a long time and costs a lot.At the same time,it will directly affect the production efficiency affected by the measuring environment.Based on the above background,the introduction of a photogrammetry system on the track board measurement and the use of automated control methods make the measurement process simpler and more efficient.Basic photogrammetry is based on three-dimensional reconstruction modeling measurements.The image-based 3D reconstruction technology includes image processing,pattern recognition,machine learning,and other related technologies.By moving the camera,different image sequences of the object are acquired at different positions,and then image transformation,feature point mapping,point cloud construction,and texture mapping are performed.Finally,the target three-dimensional model is simulated through computer programming.Based on the research of image 3D reconstruction,the main problem is the accuracy of the final reconstructed 3D model,especially in the field of industrial measurement.At present,there are many excellent results based on image-based 3D reconstruction.The three-dimensional reconstruction process is complex,and each process determines the final three-dimensional structure.This paper focuses on the basic process in the three-dimensional reconstruction process,researches the threedimensional reconstruction of multiple images in a large space,and completes the design and implementation of the detection system of the track plate geometry based on multiple image reconstruction techniques.Thesis includes the following:First,the article first elaborated the related technologies in the three-dimensional reconstruction work,and related concepts such as the camera model and the basic matrix.This paper analyzes the SIFT feature point matching algorithm,proposes the method used when the feature matching has a mismatch,and introduces a solution to the efficiency problem of a large-scale image for detection and matching.Based on this,a higher performance and high accuracy feature matching algorithm is proposed.Secondly,using the theory of object detection in images,the images taken from multiple angles will give a panoramic view of the orbital plate and detect the location data of the pre-embedded casing in the image of the orbital plate.The key technologies are image splicing and statistical machine learning,which mainly include the splicing of the orbital plates,and then use AdaBoost to detect the position data of the preembedded casing in the image.Third,the basic steps of reconstruction based on image sequences are summarized,and the projection relationship between the two images is analyzed.According to the matched pair obtained by feature matching,this kind of mapping relationship can be obtained,and then three-dimensional information of feature points can be restored to obtain the initial model.On this basis,new images are added for continuous iteration,and finally a three-dimensional model of the target is obtained.Fourth,based on the theory of this thesis,a simple 3D reconstruction system is realized.The relevant parameters of the track slab can be detected by the system.In this paper,the position information of the pre-embedded casing of the track slab is detected to verify whether the center distance of the pre-embedded casing meets the standard and whether the quality of the final decision track slab is qualified.