| Structured light 3D reconstruction technology is an interdisciplinary technology involving digital image processing,optics,automatic control,hardware and other disciplines.At the same time,as a non-contact,high-precision and efficient reconstruction method,it is widely used in industrial detection,3D modeling,VR/AR,biomedicine and other fields.When there are significant fluctuations or reflections on the surface of an object,both monocular and traditional binocular systems cannot fully reconstruct the object.In this regard,this article improves the traditional binocular system by dividing it into two monocular subsystems,left and right,by reusing projectors.The point clouds generated by two monocular systems can complement each other in the concatenated field of view,effectively solving the problem of field occlusion and reconstruction of reflective object surfaces.However,this introduces a new problem: due to certain differences in reconstruction accuracy between the two subsystems,ghosting occurs in the common area after point cloud stitching,and there is a certain sense of paragraph at the junction of the two point clouds.In this regard,this article focuses on in-depth exploration of system calibration,stereo matching,point cloud stitching,and other aspects.The main research content of the thesis includes the following aspects:(1)The traditional calibration method based on reprojection error cannot reflect the difference in reconstruction accuracy between two subsystems.This article proposes a system calibration method based on 3D reconstruction error to improve the consistency between the two subsystems.This method uses the results of Zhang’s calibration method as the initial parameters to perform 3D reconstruction on the calibration points,and optimizes the parameters of the participating system in the projector using the objective function proposed in this thesis.Experiments have shown that this method can reduce the difference in reconstruction accuracy between two systems while ensuring accuracy.(2)Aiming at the problem of low monocular point cloud density caused by the lower resolution of the projector than that of the camera.This thesis proposes a sub-pixel matching algorithm based on binary search and interpolation.In the algorithm,the bisection method is used to search for interpolable positions in the projector matrix,and then the number of matching points is increased by linear interpolation.Experiments show that the average matching time of the system in this thesis is only 0.5s,and the average reconstruction efficiency is about 21600points/s;aiming at the problem that the difference in reconstruction accuracy between the two monocular systems in this thesis cannot be completely eliminated.Firstly,the limitation of point cloud registration based on the traditional ICP(Iterative Closest Point)algorithm is analyzed.That is,the traditional ICP algorithm will evenly distribute the registration error to the entire point cloud.Since the common area of the two monocular point clouds in this thesis is large,the proportion of the point cloud area in the details that require field of view compensation is not high,resulting in this part.Regions cannot achieve good registration results.In order to solve this problem,this thesis proposes a point cloud stitching algorithm based on local registration compensation.Based on the left monocular point cloud,the right monocular point cloud is used to register the holes and missing parts in the left monocular point cloud.compensate.By increasing the "weight" of the area to be compensated,the algorithm makes the point cloud at the details have a better registration effect.Through the experimental comparison with the traditional ICP algorithm,the method in this thesis can better solve the problem of ghosting in the public area after point cloud stitching and the unsatisfactory effect of point cloud stitching in details.(3)Based on the Qt platform,this thesis develops a set of 3D reconstruction software based on structured light,which integrates monocular reconstruction algorithm,traditional binocular reconstruction algorithm and the reconstruction algorithm used in this thesis.And this thesis makes a modular design for it,so that the software can independently complete tasks such as camera calibration,projector calibration,scanning objects,and point cloud visualization. |
