3D Reconstruction Of Workpieces Based On Structured Light And Auto-encoding Learning

In the process of modern industrial production,the requirements for the precision of the workpiece are getting higher and higher.However,the traditional manual measurement and three-coordinate measurement methods are difficult to measure the contour of the workpieces accurately.Therefore,due to the characteristics of high precision,easy expansion,and high reliability,the 3D reconstruction algorithm with structured light have gradually become the research hotspots in the field of 3D reconstruction of workpieces.in this case,we adopt the principle of binocular vision and the structured grating projection algorithm to study the structured light three-dimensional reconstruction algorithm.At the same time,in order to facilitate obtaining the final mesh model of the workpieces,we have also done some researchs on the 3D reconstruction technology based on deep learning,and directly outputs the surface mesh model of the object from the network model.The main research contents of this thesis include:(1)The basic theory of structured light three-dimensional reconstruction and the specific realization process of the gray code combined with phase-shifted coded structured light algorithm are studied,and the reason for the error caused by the phase unwrapping of the algorithm is analyzed,then a signal synchronization-based method is proposed.The phase unwrapping algorithm simplifies the phase unwrapping steps,improves the adaptability of the algorithm and the point cloud calculation accuracy of the object,providing accurate point cloud data for the subsequent generation of the grid model.(2)The principle of 3D reconstruction auto-encoding network model is studied,and the advantages and disadvantages of the three 3D object representation forms,voxel,point cloud and grid,are analyzed.Based on the non-parameter auto-encoding 3D reconstruction network,combined with the manifold structure,a new parametric auto-encoding network for 3D surface reconstruction is proposed.The network can obtain the surface mesh model of the object from the input point cloud and solve the problem that the existing network can't extract the local features of the point cloud.(3)An experimental platform for structured light 3D reconstruction was built,and the camera and projector were calibrated using Zhang's calibration method.Based on the phase unwrapping algorithm and the network model proposed in section 3 and section 4,using pipe workpieces in different shapes as experimental objects,and efficiently complete phase unwrapping,stereo matching,point cloud computing and mesh generation process.verify the validity of the algorithm in this paper,and thus set up a complete set of the workpiece surface 3D measurement and reconstruction of the verification system.