Research On Texture Mapping Method Of 3D Model Based On Multi-view Images

Cultural relics are important materials for the study of history,culture and politics.However,due to the influence of human activities and natural factors,many cultural relics are constantly being destroyed.Therefore,digitization of cultural relics can be preserved permanently,which is of great significance and application value in the field of conservation and restoration of cultural relics.This paper aimed at the present problems of poor reconstruction effect,high cost and low efficiency in digital technology,puts forward a new set of 3d model based on multi-view image texture mapping method and applied to digital protection of cultural relics.Based on the existing model's point cloud data and the multi-view high-resolution images captured by the camera,texture mapping finally obtains the high quality and high fidelity 3d cultural relic model.The main work is as follows:(1)A method of 3D model texture mapping based on multi-view images is proposed.Combined with geometric model and multi-view images,the corresponding relation between the images and the point cloud data is obtained by registration of two-dimensional image and point cloud data,and then a high-quality texture model is constructed by seamless texture mapping based on the registration results.Compared with traditional methods,this method has the characteristics of high accuracy,high efficiency and low cost.(2)Registration of multi-view images and point cloud data.Reconstructing multi-view images into a sparse point cloud,and the registration of the two dimensional images and the 3D point cloud model is converted to the registration of two three dimensional point clouds.A rough registration algorithm based on random sampling uniformity is proposed to improve the accuracy and efficiency of rough registration.Finally,the transformation matrix of two point clouds is obtained by fine registration on the basis of rough registration results,and the registration of multi-angle image and point cloud model is completed.(3)Seamless texture mapping of multi-view images.An improved Markov random field based seamless texture mapping method is proposed,and the texture data is reduced and texture gaps are processed.Algorithm analysis and experimental results show that the texture mapping effect obtained by this method is more realistic.