An image dense matching method

Abstract

The invention provides an image dense matching method comprising the steps of extracting feature point images and establishing rotation and translation-invariant feature vectors; performing same-name primary matching to obtain a maximum same-name point pair subset with a stable relative relation; building a first Delaunay triangular network of two adjacent image planes and checking whether the topological relation of each pair of same-name triangles in the two adjacent images is correct; removing all same-name points with inaccurate topological relations, building a second Delaunay triangular network and calculating affine transformation parameters; calculating each non-feature image point position of dense matching in the first image, judging the topological containment relations between each non-feature image point and triangles in the second Delaunay triangular network, and calculating multiple initial same-name image point positions of the multiple affine transformation parameters in the second image; establishing a local correlation matrix of the two adjacent images; calculating and obtaining a dense matching result; and performing three-dimensional topography reconstruction on a stereoscopic observation area.

Description

technical field [0001] The present invention relates to image three-dimensional reconstruction technology, in particular, relates to an image dense matching method. Background technique [0002] Image dense matching is the most important task in image 3D topography reconstruction of photogrammetry or computer vision, and it is also one of the core tasks in current digital city construction. Only relying on the sparse and irregular feature points in the image cannot allow the general public to understand the three-dimensional appearance of the target scene observed in the image. It must rely on the dense matching formed by the non-feature points that account for the vast majority of the image to accurately and comprehensively describe it. However, in the current methods, for three-dimensional areas with poor texture and high similarity between adjacent local areas, there is always the problem of matching non-feature points in the image, and the amount of calculation is large...

AI Technical Summary

Problems solved by technology

[0004] Aiming at the problems of large amount of matching calculation and low matching accuracy in the image intensive matching method in the prior art in the three-dimensional area with poor texture and high similarity between adjacent local areas, the purpose of the present invention is to propose an image intensive matching method that can Work in 3D areas with poor texture and high similarity between adjacent local areas, and use less computation to obtain dense matching results with higher precision levels

Images