Research And Application Of High Precision Measurement Method Based On Computer Vision

High accuracy measurement based on computer vision is one of the hot research topics in the field of computer vision.With the deepening of the theory of computer vision research and constantly improve of the imaging equipment performance,high accuracy measurement based on stereo vision measurement theory is greatly appreciated by the researchers,and has been widely used in the public security,transportation and national defense military and other fields.Based on stereo vision measurement theory,this paper aimed at the urgent need of non-contact accuracy measurement of railway infrastructure geometry,focusing on high accuracy three-dimensional measurement based on three-dimensional scene reconstruction,including key technology such as landmark position,feature point extraction and feature points three-dimensional reconstruction,and put forward rail creep displacement measurement method based on computer vision,built the rail creep displacement measurement image datasets,verified the validity of the method through a large number of field experimental data.This method can satisfy the actual demand of rail creep displacement high accuracy measurement.The main research contents and results are as follows:(1)This paper proposes a two step accurate localization method for circular coded target based on deep learning.The method is divided into two stages,which firstly uses rail creep displacement measurement image as the training set of R-FCN model to achieve the initial location of mark points,and then achieves the accurate location of mark points through the algorithm of precise location of mark points based on gray centroid.Experiments on the scene image dataset show that the proposed algorithm significantly improves the detection accuracy and localization accuracy of small targets in complex scenes,and lays the foundation for the next step to detect and identify landmark matching points.(2)This paper presents a new algorithm for extracting multiple matching points based on the feature of spatial structure.The method can not only identify the center point of the mark,but also encode the diagonal points according to the spatial structure of the detected sub-pixel corner points.Thus,the number of matching points on a marked point can be expanded by 4-28 times,which provides more matching points for stereo matching between images.The robustness and effectiveness of the proposed method are verified by experiments.(3)A new method for measuring the creeping displacement of rail based on 3D reconstruction of feature points is proposed.According to the matching feature points,the 8 point method is used to estimate the fundamental matrix,and the robustness of the estimation is improved by using the uniform random sampling RANSAC method based on region partition in this paper.We also use the characteristic matrixto calculate the motion matrix of the camera;recover the 3D coordinates of the feature points by triangulation according to the camera motion matrix;use the method of beam adjustment to optimize the camera motion matrix and the spatial coordinates of the feature points in three dimensional space;reconstruct the feature points in different time periods into a coordinate system to realize the displacement measurement.Finally,the effectiveness of the method is verified by experiments.In this paper,the proposed method is tested on the data of the rail crawling scene images collected at different time points of the Beijing Kowloon line.The experimental results show that the proposed method has good robustness,high accuracy and efficiency.