Research On 3D Image Reconstruction Based On Deep Learnin

With the continuous development of technology in the field of computer vision,more and more research is devoted to 3D images.Compared with the two-dimensional image information,the three-dimensional model is more realistic and can show more information to people.Therefore,the research of three-dimensional image technology has become a research hotspot in the field of computer vision in recent years.This thesis studies some popular neural networks such as convolutional networks and residual networks,analyzes the limitations of dense point cloud reconstruction network,i.e.DensePCR,and a corresponding solution is provided at last.The main research work is as follows:1.Propose a three-dimensional dense point cloud reconstruction method based on DensePCR(Dense 3D Point Cloud Reconstruction)and pyramid architecture,namely DensePCR++,by adding multiple residual modules to global feature extraction,upsampling local features,and introducing feature fusion.Experiments show that this method can improve the object detail reconstruction effect while maintaining the point cloud resolution.The 3D image reconstruction effect is close to the real sampling model,and the performance improvement is 5.7%,which is faster than DensePCR.2.A three-dimensional reconstruction post-processing moving cube algorithm based on adaptive moving equivalent points and bilateral filters is proposed.Experiments show that this method can speed up the calculation time and reduce redundancy.The average time is reduced by 39%,and the average time is reduced by 49%.The surface reconstruction mesh is optimized,combined with the CGAL geometric algorithm,which effectively improves the performance of the mesh model with holes on the surface.3.Combined with actual application,design and implement a 3D image reconstruction software system.The main modules include:single image 3D reconstruction and 3D reconstruction post-processing.The developing process of this system is scheduled in detail,and the software is tested and verified as required.