Incremental SFM 3D reconstruction based on monocular

Abstract

When using images for 3D reconstruction, the accuracy of feature matching is a very critical. The features of image extraction and the results after matching will directly determine whether the camera pose estimation is reliable. First, in order to get accurate poses, a mismatching filtering algorithm based on local correlation of images is proposed. To make increase the number of matches, SIFT and ORB feature matching are merged as inputs to sparse reconstruction. Then use incremental SFM algorithm to get sparse 3D points from the picture set. Finally use the combination of optical flow and ORB features to densely reconstruct the image. It has been proved by experiments that when filtering and fusion-matched results are used, the number of iterations can be effectively reduced in the BA solution stage. The improved dense reconstruction algorithm can reduce the reconstruction time while ensuring the reconstruction visual effect. (Abstract)