Binocular Vision 3D Reconstruction Method For Machine-sprayed Sheet Surfaces

Automatic spraying system of furniture board usually uses the sweep spraying method to spray the board,which will lead to uneven spraying on the groove area of the furniture board surface.To solve this problem,the spatial texture of the board surface is obtained by using the technology of three-dimensional reconstruction,which planning the spray path and repairing the groove to improve the uniformity through multi-directional spraying operation.The 3D reconstruction based on binocular vision are more real-time and efficient than the laser 3D scanning technology in engineering applications,for which the 3D reconstruction method based on binocular vision theory for wooden boards is proposed.This article focuses on image enhancement in high-frequency areas,feature extracting and matching,eliminating mismatches method based on fusing algorithm of RANSAC and distance ratio criterion,and the study of reconstructing and repairing for 3D point cloud.The main work is as follows:(1)The image acquisition system and experimental platform are designed based on parallel binocular according to the actual working conditions.Then the experiment of camera calibration is carried out and the calibration parameters are verified for reliability.(2)After the feature points of binocular image are extracted and matching,the rotation matrix and translation matrix are calculated by the normalized eight point method,and the sparse 3D texture of experimental plates are reconstructed by triangulation method.Through the analysis of point cloud,the problems of point cloud missing and space deformation need to further addressing.(3)In order to solve the problem of point cloud deformation caused by mismatching seriously on feature point,the principle of Shi-Tomasi feature point extraction algorithm and Kanade Lucas Tomasi(KLT)based feature point matching algorithm are studied.Comparing the corner matching accuracy of various types of image high-frequency region enhancement methods,the experimental result shows that the Shi-Tomasi feature point extraction algorithm merging with the high-frequency image enhancement is more suitable for this project,which improves the feature point matching rate from the original 86.95% to more than 90%.Among them,the feature points matching rate of high-frequency enhancement fusing Scharr operator is the highest,94.17%.(4)The holes are located and the corresponding image blocks are extracted by searching in different regions from the original data.The new matching points and the original matching points are combined into a new set of matching points by secondary matching using reducing the local matching standard,which result in increasing the local mismatches.Therefore,the mismatching points are eliminated by combining RANSAC and distance ratio algorithm.Selecting the ratio of the eliminated mismatches to the total matched points as the performance evaluation standard.In particular,the elimination rate of fusion algorithm is 12.80%,which higher than the RANSAC method(6.91%)and distance ratio algorithm(7.01%).The sparse matching points are used as seed points for growing the regions,and then the dense 3D point cloud of wood plate is reconstructed.At last,the 3D reconstruction effect of the algorithm is verified by multiple sets of experiments(board specification: length 2050 mm,width 860 mm,texture depth 20mm).It shows that the average time of 3D reconstruction is 130 s ± 15 s,the error of inner texture depth is 2.8mm.The effectiveness and performance of 3D reconstruction meets the needs of the project and has potential value for engineering application.