Design And Implementation Of Three-dimensional Reconstruction Algorithm Based On Two-dimensional Video

Three-dimensional reconstruction technology,as an important part of computer vision research,can directly imitate human vision to process scene information.Compared with two-dimensional images,it can obtain more intuitive visual effects.The traditional three-dimensional reconstruction method mainly uses multi-angle shooting images for three-dimensional modeling,which is complex and has strict requirements for shooting angle.Monocular continuous video contains a lot of depth information and many ways to obtain it.Using monocular video information can also realize three-dimensional reconstruction.This paper designs and studies the algorithm of three-dimensional scene reconstruction using monocular continuous video.The main contents are as follows:1.A video image of key frame extraction algorithm is designed,which uses an improved method combining shallow extraction and deep filtering to process the video.Firstly,the key frame sampling algorithm is used to process the video,and the image information in the video is extracted.After getting the initial key frame sequence,the hierarchical extraction method is used to further extract the frame image.The results show that the key frame extraction algorithm can effectively extract the image sequence for three-dimensional reconstruction.2.A Harris-SIFT feature point extraction algorithm is designed.Firstly,the key points are generated by Harris corner detection algorithm,and then the descriptors are calculated by SIFT algorithm.In order to avoid the mismatch caused by matching feature point,the KLT algorithm is used to track the feature points,and the camera parameters and sparse point cloud model are obtained.3.Plane sweep algorithm and geometric guidance constraint algorithm are used to calculate the depth map of image sequence,and the results are evaluated by structural similarity algorithm.At the same time,interpolation filling,connected component and threshold elimination are used to optimize the depth map.The method based on the test between adjacent depth maps is used to fuse the depth maps,which effectively eliminates the redundant information between point clouds.Verified by real scene data and public database,this method can reconstruct realistic three-dimensional scene model.The reconstruction accuracy is better than the classical reconstruction algorithm,and the surface hole is less.