Method for purifying error matching of visual image characteristic points based on ORB (Oriented FAST and Rotated BRIEF)

Abstract

The invention relates to a method for purifying error matching of visual image characteristic points based on ORB (Oriented FAST and Rotated BRIEF). The method comprises the following steps of readingtwo to-be-detected left and right images shot under different viewing angles, constructing scale image pyramids for the images and performing grid processing; detecting characteristic points in eachsmall grid of each layer of pyramid images, extracting the characteristic points and determining a characteristic point coordinate; removing the characteristic points close to edges of the images, andcalculating the direction of a barycenter of the remaining characteristic points; calculating descriptors of ORB characteristic points; performing rough matching on the characteristic points of the two images; screening rough matching pairs of the characteristic points; removing error matching pairs again; and performing an RANSAC (Random Sample Consensus) algorithm iteration on the remaining characteristic points, and outputting the purified matching images. According to the method for purifying error matching of the visual image characteristic points based on the ORB, the detected characteristic points are uniformly distributed, the characteristic point clustering effect caused by easy huddle of multiple characteristic points is avoided, and the matching speed is accelerated while improving the matching accuracy.

Description

technical field [0001] The invention belongs to the fields of computer vision and image processing, and in particular relates to an ORB-based method for purifying incorrect matching of visual image feature points. Background technique [0002] Feature detection, description, and matching technology is one of the key steps to realize image fusion, image correction, image mosaic, and target recognition and tracking. It is also a research hotspot in the fields of image processing and machine vision navigation. The realization of many technologies such as image recognition, video tracking, image stitching, and 3D reconstruction depends on the detection, description, and matching of image feature points. High-accuracy image matching is the condition and key to determining the robot's motion. In visual SLAM based on feature points, there are often a lot of mismatching information in the feature matching results, resulting in low accuracy of the pose obtained by calculation and pro...

AI Technical Summary

Problems solved by technology

The ORB algorithm has fast operation speed, good illumination robustness, and good affine performance, but it does not have scale invariance, which leads to limitations in the application field of ORB.

[0004] In the process of image matching, since the extraction process of feature points and the description of feature points cannot be absolutely accurate, there are many wrong matches in the obtained initial matching pairs. The wrong matches are mainly divided into two situations: one is the matching The feature points are wrong, and the second is that the feature points on the image cannot be matched

The RANSAC algorithm is a classic method for eliminating feature mismatches. It has the advantages of high matching accuracy, high reliability, and strong robustness. All matched feature point pairs are iterated, and when there are many mismatched pairs, a lot of time is wasted

Images