Research On 3D Reconstruction Method Based On Radial Imaging

The reconstruction of 3D(three-dimensional)structures of objects based on two-dimensional images is an important branch of computer vision,and it has been more and more widely used in many fields.The traditional 3D reconstruction technology mostly uses two or more two-dimensional images to restore the 3D structure of the object.This article based on the radial imaging system,studies the method of reconstructing 3D structure based on a single image,focusing on the key technologies such as the basic theory of the radial imaging system,3D reconstruction methods and building experimental platforms.The imaging principle of radial imaging system,as a theoretical basis for 3D reconstruction,this article analyzes it in detail.Based on the radial section diagram of radial imaging system,this article analyzes the performance characteristics of radial imaging systems and studies the trajectory of the virtual viewpoints of camera.For the radial imaging system with different parameters,the position of the binocular space and the trinocular space in the system was deduced,and the resolution of the system was analyzed,especially the tangential resolution.In addition,the 3D radial imaging system is simulated using simulation software and some simulated radial images are obtained.For the 3D reconstruction,this article introduces and analyzes the key steps of the reconstruction method in detail,and the processing result of each steps are provided based on the simulation radial image.The method of 3D reconstruction includes four steps.Firstly,performing horizontal stacking of radial images using bilinear interpolation method.Secondly,performing stereo matching on images corresponding to different viewpoints,comparing the two methods of SIFT(Scale Invariant Feature Transform)feature matching and area matching,the latter is more conducive to obtain dense matching point pairs,especially for the texture of target is not rich enough,the latter method is better.In addition,according to the imaging principle of radial imaging system,this article deduce the calculation formula for depth of the space point,so that the depth information of the space point is calculated with the position of the matching point pair,and the 3D coordinates of the space point can be obtained with the line coordinates of the point in the horizontal stacking diagram.The 3D point cloud can be obtained by calculating the 3D coordinates of all matching point pair.Finally,based on the 3D point cloud,the 3D model of the target is reconstructed using the Poisson reconstruction method.Finally,the experimental platform for radial image is built,and 3D reconstruction based on the real-shot radial image is completed,which verify the feasibility of the 3D reconstruction based on the radial imaging system.In the of process experimental platform construction,some key issues were analyzed and discussed,such as the alignment of the camera optical axis and the internal mirror,and the performance requirements of the camera in the radial imaging system.At the same time,the tangential resolution of the real radial image is analyzed to verify the theoretical research on the tangential resolution.In order to evaluate the reconstruction accuracy of the system,the target of the known geometric shape is reconstructed,and the point cloud obtained by the 3D reconstruction is surface-fitted,the reconstruction error is obtained by comparing the surface of fitting and the real surface of target,the result show that the error is within the acceptable range.Finally,extending radial imaging system to apply to reconstruct the complete geometry of the target.