Fast iterative computation method for semi-dense stereo matching

Abstract

The invention relates to the field of computer vision technologies, and discloses a fast iterative computation method for semi-dense stereo matching, which accelerates the matching speed while satisfying the requirement of acquiring high matching precision. The fast iterative computation method can be summarized in the steps of: extracting feature points of a left image and a right image separately, and constructing feature descriptors; then, completing feature point matching of the left and right images according to the feature descriptors and an epipolar constraint, wherein the successfullymatched feature points are called supporting points; further, constructing a Delaunay triangle according to the supporting points in the left image and estimating prior parallaxes d of all pixel points in the left image; and calculating cost errors C corresponding to the prior parallaxes d, and acquiring the minimum cost error C<min> of all the supporting points during current iteration;iteratively updating a supporting point set and the minimum cost error until an iteration termination condition is satisfied; and finally, calculating a matching point in the right image of each pixelpoint in the left image according to the parallax of the pixel point.

Description

technical field [0001] The invention relates to the technical field of computer vision, in particular to a fast iterative calculation method for semi-dense stereo matching. Background technique [0002] Image stereo matching is an important branch of computer vision, photogrammetry, and computer graphics, and it is of great value in many applications. Image matching can be divided into sparse matching and dense matching. Sparse matching generally extracts the feature points with strong texture on the image, and then calculates the matching cost through the feature descriptor to obtain the optimal matching. Due to the sparsity of feature points, sparse matching cannot provide a sufficient number of feature points and 3D points in many applications, so only limited 3D world information can be obtained. Dense matching is to match each pixel of the image, so dense 3D world information can be obtained. Algorithms for dense matching can be divided into global methods and local ...

AI Technical Summary

Problems solved by technology

However, due to its inherent computational limitations, dense matching cannot achieve higher processing frame rates (eg, VGA image efficiency of 30fps or higher under equivalent computing conditions)

[0003] Therefore, the above two stereo schemes in the traditional technology have the following defects: sparse matching has high processing efficiency, but cannot provide a sufficient amount of matching point information; dense matching can obtain dense matching point information, but cannot obtain high processing efficiency , unable to adapt to the increasingly high industry requirements

Images