A Structure-Adaptive View-Invariant Linear Feature Matching Method for Wide Baseline Images

Abstract

The invention discloses a structure-adaptive wide-baseline image perspective-invariant line feature matching method, which belongs to the field of remote sensing image processing. The method includes extracting reference image straight line features and search image straight line features from the reference image and search image respectively and compiling multiple reference image straight line feature groups and multiple search image straight line feature groups; The feature group performs the initial matching of the straight line features of the angle of view invariance, and estimates the epipolar geometric relationship between the reference image and the search image; The second matching is to perform single-line feature matching on the reference image line features and the search image line features that have not been successfully matched; the results of all three matchings are combined to obtain the final matching result. The present invention proposes a straight line feature matching method with constant viewing angle, which effectively improves the number of matched straight line features and the correct rate.

Description

technical field [0001] The invention belongs to the field of remote sensing image processing, and in particular relates to a structure-adaptive wide-baseline image viewing angle-invariant line feature matching method. Background technique [0002] With the continuous improvement of image resolution, most of the artificial objects on the image, such as buildings and roads, can be described by straight line features. The straight line feature is located on the edge of the target contour, contains rich semantic information and geometric meaning, can better express the target structure, reflect the high-level information of the image and the target scene, and use the straight line feature as a matching primitive for feature matching in fine 3D modeling of images, etc. important in application. [0003] After years of development, researchers have proposed a large number of line feature matching methods suitable for different scenarios. According to the difference of constraint...

AI Technical Summary

Problems solved by technology

The above methods based on feature descriptors have a common deficiency: when there is a large change in viewing angle between the stereo pairs and the depth of the object in the image area changes significantly, the regular feature areas in these methods and the feature descriptions built on this basis The robustness of symbols is poor, which affects the final matching results

However, in this method, the symmetrical feature area constructed around the intersection point of the line feature pair is difficult to adapt to the geometric deformation of the image area with discontinuous parallax on the wide baseline image (such as the edge of the building, etc.)

Part of the image content in the feature area is difficult to maintain consistency under the change of image viewing angle, resulting in low similarity of feature descriptors with the same name

In addition, the method of finding salient points on straight line features is susceptible to the inconsistency of the lengths of straight lines with the same name, resulting in unreliable feature descriptors constructed

Images