Research On 3D Reconstruction Algorithm Based On Image Data

With the rise of artificial intelligence presently,computer vision has become very popular as an important branch of artificial intelligence,and three-dimensional reconstruction is one of the more popular research in this field.The main research content of three-dimensional(3D)reconstruction is how to obtain 3D information by reconstructing a single view or multi-view.This paper mainly studies 3D reconstruction from two aspects,which are 3D reconstruction algorithm based on motion recovery structure and 3D reconstruction algorithm based on deformable model,and make some related improvements.The main work of this paper is as follows:1.The related technical principles are also introduced,such as projective geometry,camera model and imaging principle,pole geometry,base matrix,feature point detection and matching,3D deformable model foundation and so on.2.The 3D reconstruction algorithm based on motion recovery structure mainly introduces the whole process of 3D reconstruction of multi-frame disordered images,including the basic principle of camera calibration,key frame screening of disordered images,matching algorithm of feature points,sparse reconstruction of 3D objects and dense reconstruction of three-dimensional objects.Since there is no mature method to effectively select the key images from the disordered images for 3D reconstruction,this paper presents a method based on DDCRP clustering algorithm to solve this problem.In this method,the key words in the sequence of the unordered images are sorted by clustering the BoW word package model of the image.It can improve the efficiency of reconstruction under the premise of maintaining the result of reconstruction.Through the experiments on multiple sets of data sets,the results show that the reconstruction algorithm is highly efficient and available.3.The 3D reconstruction algorithm based on deformable model mainly introduces the calculation and attitude estimation of shape model,texture and light modeling.This algorithm uses a picture for 3D reconstruction.Based on the 3D average model of known objects,the algorithm can capture the two-dimensional image and its specific feature points of the corresponding model,and can iterate until the convergence is completed.During the iterative process of the algorithm,this paper introduces the probabilistic principal component analysis method to optimize the reconstruction details on the basis of the traditional principal component analysis method.Since the reconstruction algorithm based on motion recovery structure cannot reconstruct the face data set well in the experiment,the 3D reconstruction experiment based on the deformable model is carried out,and the results of reconstruction show that the proposed method is efficient and feasible.In this paper,two sets of reconstruction algorithms are used to reconstruct the common data sets.The reconstruction results show that the algorithms are efficient and practical,and can satisfy our specific requirements.