Research On 3D Reconstruction Technology Of Ancient Relics Based On Single View And Deep Learning

Ancient relics carry the historical,cultural and spiritual connotation of human civilization for thousands of years,and play a vital role in the study of ancient economy and civilization.Through the digital processing of ancient relics,the transformation from two-dimensional images to three-dimensional models can effectively reduce the loss of space and time resources brought by physical exhibitions.At the same time,it can enable people to more comprehensively understand and learn the history and culture contained in ancient cultural relics from a three-dimensional perspective through the form of network.Different from the traditional 3D reconstruction methods,the 3D reconstruction method based on deep learning is more suitable for the case of single view,and can carry out 3D reconstruction without complex image processing operations.Therefore,it is of far-reaching significance to study the 3D reconstruction technology of antiquities based on single view and deep learning.The specific research contents of this paper are as follows:Aiming at the problem of insufficient extraction ability of current feature extraction algorithms,a 2D high-order module is proposed by using the existing high-order information extraction algorithms.Based on the existing data sets of cultural relics and handicrafts,the 2D high-order module is realized by 1 × 1 convolution and operator quadrature and summation at the encoder level.Using this module,we can mine the depth information of image data,model the complex high-order relationship in the picture,improve the performance and effect of feature extraction,and solve the problems of insufficient extraction depth,rough and unclear features in the first-order feature extraction.The IOU index of using 2D high-order module combined with RESNET as encoder is 0.570,which is 1.2% higher than that of 3d-r2n2 using RESNET only as encoder.Aiming at the problem of insufficient accuracy of existing reconstruction algorithms,using "encoder decoder optimizer" as the overall network structure,a 2D repvgg(2D re parameterization visual geometry group)network model is proposed as the encoder.The model simulates multi branch parameters in the form of single branch,which has better accuracy and performance than the residual network.Am softmax is used as the loss function,And thhe context aware fusion module is used as the optimizer,which can optimize the details of the voxel model.The experimental results show that the IOU value of the overall network model is improved by 0.7% compared with pixvox.According to the above research contents,this study designs and implements a 3D reconstruction system of ancient relics based on single view and 3D reconstruction.The system can realize the functions of image upload,three-dimensional reconstruction and data download.For the image data input by the user,the system can automatically detect the image and generate the corresponding three-dimensional voxel model when the input conditions are met.Experiments show that the whole system is efficient and available,and all functions can be performed correctly.