Research On 3D Reconstruction Of Large Scene Based On Unordered Images

Multi-view geometric 3D reconstruction has been a hot topic in computer vision.It has a wide range of applications in the fields of visual navigation,medical diagnosis,virtual reality and so on.The incremental 3D reconstruction algorithm based on the structure of motion is mature and can be used to construct large scene 3D model.However,the algorithm is not accurate enough for unordered image matching,and the algorithm takes a long time.In this paper,a large scene 3D reconstruction algorithm based on unordered images is studied,and a global 3D reconstruction algorithm based on AKAZE feature extraction is proposed,which has the advantages of high matching accuracy,small point cloud error and shorter time cost.The specific work of this paper is as follows:1.The feature point extraction and matching algorithms are studied,SIFT algorithm and AKAZE algorithm are discussed,and the K-D tree of data structure used in feature matching is introduced.Aiming at the problem that the K-D tree needs to trace back to the root node for each matching feature to end,the paper proposes to sort the nodes on the query path to match from the highest priority node to improve the matching efficiency effectively.2.The algorithm of constructing camera pose matrix is studied,and the classical method of constructing image association graph from second view is introduced.This paper presents a method of constructing unordered association graph by using three view constraints,which makes the correlation between images more closely.It can solve the global rotation and translation matrix and provide better input information for the 3D reconstruction.3.The 3D reconstruction algorithm of unordered image is studied,and the traditional incremental 3D reconstruction algorithm is introduced.Aiming at the problem of increasing the error accumulation caused by incremental 3D reconstruction iteration,a global 3D reconstruction algorithm based on AKAZE algorithm is proposed.Only one cluster adjustment is needed to solve the problem of error accumulation in incremental reconstruction.At the end of this paper,the experimental results of global 3D reconstruction algorithm based on AKAZE algorithm and incremental 3D reconstruction algorithm are given.The experimental results show that the proposed method can achieve satisfactory reconstruction results.The feasibility and superiority of this method are verified.