Research On 3D Reconstruction Of Mine Environment Based On Machine Vision

With the rapid development of various fields in China,as the main energy,the demand for coal is increasing year by year.The coal industry has thoroughly implemented the national energy security strategy and gradually promoted the development of unmanned and intelligent coal mining.The operation environment of underground coal mine is very special,which is a complex terrain environment mixed with structured and unstructured,and the mine scene after disaster belongs to unstructured environment completely.In order to ensure the daily work of intelligent production robot in coal mine and the smooth rescue task of post-disaster rescue robot,it is urgent to reconstruct the 3D environment of restricted scenes in underground coal mine.In view of the above problems,the main research contents are as follows:(1)Study the transformation relationship between 3D rigid body motion and key coordinate system.Aiming at the influence of camera distortion on image data,zhang Zhengyou calibration method and MATLAB calibration tool were used to carry out calibration experiment,and the internal parameters and distortion coefficient of camera were calculated,which reduced the influence of camera distortion on image data.(2)In view of the fuzzy problem of color images collected due to low illumination and dust and fog in the mine environment,the mainstream image defogging algorithm is compared and analyzed,and the image defogging is processed based on the optimized contrast enhancement algorithm.Aiming at the problem of void in depth images collected by Kinect2 camera,an improved FMM depth map void repair algorithm was proposed.The confidence factor is introduced and the original repair neighborhood is replaced by the fixed scale neighborhood.When the confidence is low,the search range of candidate blocks is determined according to the continuity of image features,and the similarity measure function is used to obtain the best matching block,and the center pixel value of the best matching block is replaced to the point to be repaired to complete the repair.Experimental results show that compared with the traditional FMM algorithm,the PSNR of the image processed by the proposed algorithm is improved by 11.79%,and the cavity area can be completely repaired to avoid edge distortion.(3)In view of the large amount of point cloud data to be processed and the unstable geometric characteristics of random sampling points,a two-step point cloud registration algorithm based on improved sampling consistency and ICP is proposed.Firstly,the collected color image and depth image are aligned to generate 3D point cloud and the point cloud is preprocessed.Then,the mean curvature is used to simplify the point cloud data,and combined with the two constraints of distance threshold and average density of point cloud,the feature point set and matching point set with more geometric features are generated,and the transformation relation of corresponding points among the point sets is calculated for point cloud rough registration.Finally,the point cloud transformation relation output by the improved algorithm is used as the initial value of ICP algorithm for precise registration,and a two-step point cloud registration algorithm is formed.The experimental results show that the two-step registration algorithm has high point cloud registration accuracy.(4)The feasibility study of the 3D reconstruction system of mine environment based on machine vision was carried out.The unstructured simulation environment of mine and the simulation environment of mine roadway were built in the laboratory,and the 3D reconstruction and accuracy analysis of the two environments were carried out respectively.The experimental results show that the 3d model reconstructed for the mine simulation environment conforms to the real scene,and the reconstruction accuracy can reach cm level,which verifies the feasibility of the system in the special mine environment.There are 64 figures,13 tables and 88 references in this thesis.