Image matching method

Abstract

The invention provides an image matching method, which comprises the following steps of: extracting a first point feature and a first line feature of a first image, and extracting a second point feature and a second line feature of a second image; obtaining a first line feature matching pair and a second line feature matching pair based on the first line feature and the second line feature; obtaining a first point feature matching pair and a second point feature matching pair based on the first line feature matching pair, the first point feature and the second point feature; taking two pairs of second point feature matching pairs as a group, constructing a first virtual line feature by taking two first point features as end points, constructing a second virtual line feature by taking two second point features as end points, and obtaining a virtual line feature matching pair; and constructing a similarity measurement matrix of the virtual line feature matching pair and the second line feature matching pair, and solving the similarity measurement matrix to obtain an image matching result.

Description

technical field [0001] The present disclosure relates to the fields of computer vision and robot technology, and in particular to an image matching method. Background technique [0002] Simultaneous Localization and Mapping (SLAM) technology based on visual sensors is currently a hotspot in the research and application of computer vision and robotics, and has great practical value and broad development prospects. Existing visual SLAM methods can be divided into feature methods based on feature extraction and matching and direct methods based on photometric consistency assumptions. Because the point features in the image are easy to track, and the corresponding parameterization and geometric calculation are relatively simple, the method based on point feature extraction and matching is currently a relatively common visual SLAM solution. The classic point feature method framework extracts the key points in the image through the point feature detection algorithm, further uses ...

AI Technical Summary

Problems solved by technology

[0003] However, the visual sensor that acquires image information in visual SLAM is a passive sensor that is easily disturbed by environmental factors.

The visual SLAM system based on point features relies heavily on the quality of image point feature matching in terms of positioning accuracy and system stability, and point features detected in indoor weakly textured scenes with doors, windows, long corridors, and white walls often have feature description discrimination The problem of insufficient ability, and it is difficult to detect a sufficient number of point features, which affects the positioning accuracy of the visual SLAM system and even leads to tracking failure

Compared with point features, line features can provide richer geometric information. The existing SLAM method based on point and line features extracts point features and line features to perform point feature matching and line feature matching, and then performs pose analysis based on the matching results. Calculation and optimization, however, the discriminative ability of point features is weak in weak texture scenes, and this does not make full use of the structural constraint relationship between point features and line features in the image

Images