Image feature matching method capable of resisting large geometric distortion

Abstract

The invention discloses an image feature matching method capable of resisting large geometric distortion, and the method comprises the following steps: 1), matching features of an image pair comprising a reference image and a to-be-matched image to acquire a candidate feature matching point set; 2) for the candidate feature matching point set, constructing a support line set for each matching point, constructing a feature vector through a support line feature descriptor, and performing luminosity constraint mismatching preliminary filtering based on a support line voting strategy; 3) taking the affine invariant ratio as a local neighborhood geometric constraint to carry out matching point fine filtering on a primary filtering result in the step 2); calculating an affine transformation model of a local neighborhood of each feature matching point; 4) reliable matching point expansion based on a local affine transformation model; and 5) constructing a grid affine transformation model to realize registration and splicing of large geometric distortion images. The accuracy of large geometric distortion image matching, the number of matching points and the precision of the matching pointscan be greatly improved.

Description

technical field [0001] The invention relates to photogrammetry and remote sensing technology, in particular to an image feature matching method against large geometric distortion. Background technique [0002] With the improvement of sensor and computing performance, the rapid development of photogrammetry and remote sensing technology provides strong support for disaster relief, facility construction, urban planning and other aspects. Among them, image-based aerial triangulation, 3D modeling, and target change detection are key supporting technologies for industrial applications such as surveying and surveying, geographic information system database updates, smart cities, geographic national conditions monitoring, and major disaster emergency response. The stable and reliable image feature matching method is the premise and guarantee of these key technologies, which can effectively promote the research progress of many important issues and is of great value. [0003] Featu...

AI Technical Summary

Problems solved by technology

However, traditional methods are extremely sensitive to large geometric distortions, especially non-rigid deformations

This difficult problem brings a daunting challenge to the classical matching algorithm, making it difficult to meet specific practical application requirements

Large geometric distortion brings the following specific problems to the traditional feature matching algorithm: (1) low matching accuracy; (2) difficult to eliminate wrong matches; (3) small number of matching points; (4) difficult to model geometric relationships

Images