3D Reconstruction Method Research Based On Image Sequence With Constraint Match Propagation

With the rapid development of computer vision technology, more and more attentions are paid to the acquisition of more photorealistic3D models of the scenes. Because of the high cost and restricted range of usage, traditional device based3D reconstruction method is limited to the practical application. Compared with the3D reconstruction methods based on single view or silhouette, the3D reconstruction method based on image sequences is more widely used with less constraint. With only one ordinary digital camera in demand, the3D reconstruction method based on image sequences has been one of the most popular subjects in the area of computer vision.Image matching is the crucial step of3D reconstruction method based on image sequence. Most traditional3D reconstruction methods which are based on image sequences depend on the spare feature corresponding algorithm. These spare matching points are sufficient for computing camera parameters and tracking camera positions, but not for the final3D scene structure visualization. Then, how to find more reliable matching points is worth studying. This paper does the research aimed at the match propagation problem in3D reconstruction process. The main work and contributions are given as follows:(1) This paper does the research on Quasi-dense match propagation algorithm based on neighborhood. Exploring and implementing traditional Quasi-dense match propagation algorithm. What’s more, adaptive support weight is used instead of cross correlation measurement so as to solve the mis-matching problem effectively on the image edge.(2) This paper proposes a novel match propagation algorithm based on triangular constraints. Firstly, cross-consistency check is adopted in the initial matching process in order to obtain initial matching points set which has lower mis-matching rate. Then, in order to solve the low matching score problem caused by traditional feature matching algorithm, seed point exploration based on triangle constraint is used to get more accurate seed points for subsequent match propagation. Finally, in the match propagation based on neighborhood process, triangle constraint is used, which can avoid the influence of the bad matches in each iteration process and finally get more reliable matches.(3) This paper proposes a new3D reconstruction method based on match propagation algorithm. In order to solve the problem of the bad3D model visual effect with the traditional3D reconstruction method in short or low information image sequences. Image match propagation algorithm is combined in the3D reconstruction method based on the two-view initial reconstruction. The novel method is effectively used in3D reconstruction to make the final3D reconstruction model have a better visual effect.