3D Multi-View Reconstruction And Model Re-Topology

With the development of computer vision technology,three-dimensional reconstruction technology has become a hot research direction,and it has been applied in more and more fields,such as augmented reality,historical building model modeling and so on.In the aspect of three-dimensional reconstruction,it is still difficult to generate high-quality models with the current technology.In addition,in the actual production process,the models generated by three-dimensional reconstruction have many problems such as too much data,redundant polygons or confused maps,which are difficult to edit.They cannot be directly used in augmented reality applications,3D games and other industries that require high quality models.This paper mainly improves the effect of three-dimensional reconstruction by improving the effect of feature matching and converts the generated three-dimensional mesh to semi-regular mesh by the model's re-topological algorithm.The main work of this paper are listed as follows:1.To solve the problem of poor 3-D reconstruction,we optimize the image feature matching phase in incremental sparse reconstruction.Combining the feature matching method based on grid statistics and the re-matching algorithm,the matching method based on grid statistics can exclude the mismatched feature pairs,and the re-matching algorithm verifies the candidate matching by random sampling consistency algorithm,which is realized by verifying whether there are potential matching feature points in the queue,and this method can increase the number of feature point matches,improve the effect of feature matching,and reduce the missing patches of the three-dimensional reconstruction model.2.To solve the problem that textures are difficult to process in the models generated by three-dimensional scanning or multi-view reconstruction,we propose a machine-learning based mesh re-topology algorithm.It can convert any model with texture to a semi-regular mesh,and we will retain its texture and re-process the ambient light shading on model.First,we use a neural network to find the singularity of the model,then we calculate and optimize the direction fields to generate the semi-regular mesh.Experiments show that this method can significantly reduce the number of polygons in the model,generate semi-regular models.Our method can significantly reduce the difficulty of editing model texture,and have a good topological effect.