Study Of The Improvement Matching Algorithm On Feature Points

With the continuous development of science and technology, the image as a newdata information have been widely used in various fields.In the field of visual SLAM,image feature matching technology plays an important and central role. Its primarytask is that stable feature points are extracted from the image to be matched, and thecorresponding descriptor is generated for each feature point, and then a correspondingrelationship between feature points is established through a similarity measurebetween the point and point. Because different sensor has different imaging principle,and the time, the climate and the angle of image acquisition are different, which makethe image matching increased the difficulty. The feature-based matching methodextract the salient features of the image, and thus greatly reduce the amount ofinformation in the matching process, and the characteristics of such a significanceresist some interference outside and improve the robustness of the matching.SIFT and SURF feature matching algorithm have better robustness, but for somepresence of noise or a variety of image transformation, the matching algorithm’sperformance will still be affected. Meanwhile, for some higher real-time requirements,despite SURF is an Speeded Up Robust Features, still can’t meet the real-timerequirements. Therefore, the rapid extraction of stable features, and the formation ofan effective descriptor are the main goal in this article.First of all, for the high complexity to determine the main direction of the featurepoint, and additional auxiliary direction information leading to increase the time ofthe follow-up searching the matches in the SIFT algorithm, this paper proposes tocalculate the centroid of the feature point circular neighborhood by image moments,and thus the main orientation of the feature point is the direction of the vector fromthe origin to the centroid and enhanced robustness, and speed up the execution timeof the algorithm.Secondly, aiming at the reason that SURF using Euclidean distance to dosimilarity measure is slower, and the characteristic of using integral image features, this paper proposes an improved binary descriptor by the BRIEF additional rotationalinvariance and noise immunity, and improve the SURF algorithm by Hammingdistance similarity measure to ensure the performance of the algorithm and improvethe efficiency of the algorithm.Finally, this paper describes the evaluation method for feature matchingalgorithm and introduces RANSAC algorithm used to remove mismatching, and thematching algorithms that this paper focuses on in the actual shooting of the scenemake a comprehensive comparison and analysis. The experimental results show theeffectiveness of the proposed algorithm.