Research On Point Cloud Reconstruction Based On Projected Speckle And Photograph With Single-camera

The 3D point cloud vision measurement technology in the industrial field has a wide range of application requirements in reverse engineering,product quality inspection,product shape design,mechanical parts manufacturing and so on.The traditional binocular active vision measurement method needs to project a specific pattern of structured light to the object to be measured.Besides,it is necessary to move the measuring head repeatedly to scan the object in order to obtain complete point cloud data.In this paper,combined with speckle projection technology,a point cloud measurement technology based on single camera photography is studied.The SLR camera is used to perform a global free shoting of the object,thereby the point cloud with high precision on the surface of the object is reconstructed.The main research contents and innovations of this paper include:1.A set of high quality projection speckle generation algorithms was designed.The algorithm can regulate the parameters such as the size,density,contrast and distribution of the speckle particles,and the surface texture information of the object can be effectively enhanced by the projection device.2.Aiming at the de-correlation problem of matching large visually deformed images with traditional digital image correlation(DIC)algorithm,a new RFDIC algorithm with rotation compensation is proposed.The method utilizes the rotation transformation parameters between images,and first performs rotation "rectification" on the matching image to satisfy the DIC correlation condition,then uses the DIC algorithm to perform image matching.At the same time,this paper also studies a method of Patch Match,which uses the spatial tangent plane homography model to establish the image matching relationship,and the image matching of large visual deformation is still effective.This paper also compares and verifies the performance of these two algorithms.3.A multi-view geometry solving method without coded points is proposed.In the process of solving multi-view geometry,the traditional industrial photogrammetry needs to establish the multiview geometric relationship with the help of a large number of coded points arranged on the surface of the object,so the camera pose of each shooting position can be obtained.In this paper,a method for solving high-precision camera poses without the coded point information but with only the speckle texture information is proposed.4.Based on the open source library——OpenMVG,the multi-view geometry framework is established.And the reconstruction technology of dense point cloud is realized.With the point cloud reconstruction experiment of single-camera photographing of specific industrial parts,We compared the reconstruction results with the measuring data of professional scanning equipment,and the accuracy level of the point cloud reconstructed with our method is verified.