Fast dense matching method based on close-range photogrammetry

Abstract

The invention relates to a fast dense matching method based on close-range photogrammetry. The fast dense matching method comprises the following steps of: feature extraction: an image pyramid is built for an original image, an HARRIS corner point response value is calculated for each layer of the image in the image pyramid, and the feature point direction is determined for generating feature point description; sparse matching: density reduction is carried out by aiming at feature points for forming strong feature points, and SIFT matching is carried out on the strong feature points; and dense matching: for consistent points generated by the SIFT matching, an interpolation parallax method on a plane triangular net is adopted for solving initial left and right parallax and left and right image initial matching coordinates, left image feature points are traversed for finding matching points on right images, then, right image feature points are traversed for finding matching points on left images, and the least square template matching is adopted in the matching method. When the fast dense matching method is adopted, the fast, automatic, precise and complete dense matching can be realized.

Description

technical field [0001] The invention relates to the technical field of close-range photogrammetry, in particular to a fast dense matching method for close-range photogrammetry. Background technique [0002] Close-range photogrammetry includes two processes of close-range photography and image processing. Image matching is a key technology in image processing. The so-called image matching is to find the same-name points in the two photos in the stereo pair. Traditional close-range photogrammetry uses Direct Linear Transformation (DLT) to directly solve the initial value of the external orientation elements of the image through the three-dimensional coordinates of more than six common points on the two images, and then through the polar constraint, One-dimensional image matching is performed on epipolar lines. This method requires manual measurement of many image control points, and the initial value error of the outer orientation elements is relatively large, so the matchin...

AI Technical Summary

Problems solved by technology

This method requires manual measurement of many image control points, and the initial value error of the outer orientation elements is relatively large, and the matching effect is often not ideal

In terms of matching automation, some people propose to use the HARRIS corner detection operator to extract feature points, and then use least squares template matching to match points with the same name. However, the initial matching pair often needs to manually specify the seed point, which cannot be automated.

Some people use the SIFT matching algorithm, but the feature points extracted by the SIFT algorithm are often not natural texture feature points, and the speed is very slow when matching large images, and the matched points are not dense enough.

Images