An Image Neighborhood Optimization Method for Improving the Efficiency of Large-Scale 3D Reconstruction

Abstract

The invention discloses an image neighbor optimization method of improving large-scale three-dimensional reconstruction efficiency. The image neighbor optimization method comprises steps that neighbor matching image pairs are acquired, and then the characteristic point matching of the image pairs is carried out; a fundamental matrix is used for geometric verification to eliminate error matching incapable of satisfying an epipolar constraint to acquire an interior point number, and a homography conversion rate is acquired by calculating a homography matrix; the changes of the directions and the dimensions of the matching characteristic points are counted to acquire a corresponding histogram; the similarity of the images is measured by three constraints, namely, the interior point number, the homography conversion rate, and the direction and dimension histogram, and redundant images are marked; the image pairs comprising the redundant images are eliminated, and narrow baseline image pairs are eliminated by the interior points and the homography conversion rate; and finally, image pair matching information after filtering is stored, and because the redundant image pairs and the narrow baseline image pairs are eliminated, the precision and the efficiency of the subsequent three-dimensional reconstruction are further improved.

Description

technical field [0001] The invention belongs to the field of computer vision, and more specifically relates to an image neighbor optimization method for improving the efficiency of large-scale three-dimensional reconstruction. Background technique [0002] 3D scenes of large-scale image collections have been a hot research area in recent years. The currently commonly used algorithm for 3D reconstruction is the incremental Structure from Motion (SFM) algorithm, which mainly includes the following four parts: 1) Extraction of image feature points, 2) Feature matching between images, 3) Matching The image matching pairs are geometrically verified, and 4) the camera pose and sparse 3D point cloud are estimated from the matching. For large-scale datasets, the key issue is efficiency. According to the above process, the bottleneck of the current algorithm efficiency mainly occurs in the second and third steps. The original method of the second step is to match two by two, but fo...

AI Technical Summary

Problems solved by technology

This solves the technical problems of redundancy and narrow baseline in the prior art after image neighbor search

Images