Research On 3D Reconstruction Based On AKAZE Algorithm

3D-reconstruction based on image sequence can obtain the images by shooting the specified target at different angles,selects the appropriate feature detection and matching method to obtain the position information of the target features,then calculates and eliminates the wrong pairs,and finally completes the reconstruction.In practical applications,the process of 3D reconstruction is complex,and the accuracy is affected by various factors such as the quality of the captured images and the algorithm.Feature extraction and matching is the initial and most critical part.The precision and time-consuming of feature matching greatly affects the effect of 3d reconstruction.At present,there are many kinds of feature extraction algorithms.Each of them has both advantages and disadvantages which are suitable for different scenarios.It is of great importance to choose the most suitable feature matching algorithm for 3d reconstruction.Starting from the background of topic selection,this paper reviewed a large number of domestic and foreign materials and literatures,fully demonstrated the research significance and necessity of this topic,summarized the research status and development prospect of this topic at home and abroad,and proposed a 3d reconstruction algorithm based on modified AKAZE algorithm.The specific work is as follows:1.The process of feature extraction and matching in images.The principle of classical SIFT algorithm and RANSAC algorithm are studied.Studied recent AKAZE algorithm based on nonlinear scale space.For the problem that Perona-Malik model can not filter noise well,an improvement is proposed that the Charbonnier model is used to replace the Perona-Malik model in nonlinear filter diffusion.It can filter out noise while retaining more details of the image than before.RANSAC algorithm is used to eliminate mismatch.Experiments show that the improved AKAZE algorithm improves the accuracy and speed of feature matching,which will provide more accurate match pairs for 3D reconstruction.2.The paper studied the basic concept theory of 3D reconstruction based on image sequence.The princlple of SFM to build sparse point cloud is introduced.For the problem that traditional 3D reconstruction based on SIFT algorithm will lose a lot of boundary information,this paper proposed an incremental SFM algorithm based on AKAZE algorithm.The algorithm is divided into three parts:(1)Featrure extraction and matching according to the modified AKAZE algorithm,using RANSAC to eliminate mismatch;(2)Using SFM algorithm for sparse point cloud construction;(3)Combined with PMVS algorithm,constructing a dense point cloud and realize the process of 3D reconstructon.This paper compares the result with tradtional 3D reconstruction method based on SIFT feature.The result shows that the method proposed in this paper can improve the effect of 3D reconstruction to some extent.