Reconstruction Of Manifold Surface Based On Machine Learning

Surface reconstruction is an important researcher in computer graphics and an important technology in reverse engineering.It has a wide range of applications in industrial design,film,and television,game animation,and other fields.Data structures such as lattices,voxels,and triangles.The surface constructed in this way has certain difficulties in design and modification,especially in the model redrawing requires certain design skills and art experience.The accuracy of the instrument has gradually improved,and directly using 3D point cloud data for surface reconstruction has become popular research.Due to the large amount of data based on point cloud data,but also disorder,rotation invariance and unstructured characteristics,the traditional 3D point cloud model surface and surface reconstruction based on curvature and unevenness and feature descriptors or basis functions are used The methods are difficult to ensure the arrangement and local characteristics of the model at the same time.These methods not only have a large amount of calculation but also require a large amount of prior knowledge of mathematical geometry.In order to solve the problem that the surface reconstructed with 3D point cloud data is difficult to take into account the problem of overall smoothness and local details,here we combine the existing surface reconstruction technology,the classification and segmentation algorithm of the 3D point cloud,the neural network structure and flow of machine learning based on the shaping principle,a method of manifold surface reconstruction based on machine learning is proposed.This method is a surface reconstruction algorithm based on point cloud data.The main research contents and steps here are as follows.(1)Aiming at the problem that the current point cloud segmentation cannot directly determine the depth and breadth of segmentation,a self-merging octree segmentation algorithm is proposed.The reason is that the segmentation is based on the classification method and has the same distortion The feature neighborhood point clouds are merged.Through this merge operation,the stitching problem in the surface reconstruction process can be effectively solved.(2)In combination with the existing PointNet++ point cloud segmentation classification algorithm,a point cloud segmentation algorithm of SMON is proposed here.It is a point cloud segmentation algorithm based on self-merging octrees.By constructing a surface pool to replace the mapping relationship between point clouds and surfaces,this relationship is passed to the deep learning algorithm to construct the model,and finally,one can be given to each piece.Point cloud segmentation model with point cloud labeling.This method effectively solves the problem that other point cloud segmentation algorithms rely too much on prior knowledge.(3)In the process of surface reconstruction,it is easy to appear that the model as a whole does not meet the Euclidean space characteristics,which makes it difficult to use basic basis functions to describe these surface patches so that the re-manifold principle is used as a constraint for surface reconstruction In order to ensure that the curved surface sheet has good smoothness at the initial stage of construction.Finally,these surface patches are split and fused to construct a complete model.For the holes existing in the stitching process,three B-spline techniques were used instead of repairs to ensure that the entire model was smooth after reconstruction.