基于自适应的AGAST特征均匀化提取算法

Adaptive AGAST feature homogenization extraction algorithm

特征点的提取速率和均匀程度对视觉SLAM的性能具有重要影响。为了提高传统视觉SLAM系统的特征提取速率,实现对图像特征的均匀化提取,提出了一种基于自适应阈值的AGAST特征均匀化提取算法(QOARB)。首先,通过构建图像金字塔,实现特征点的尺度不变性;其次,根据图像灰度值与方差计算AGAST角点的初始提取阈值,提高算法在不同图像上的特征提取速率;接着,用改进的四叉树方法来均匀筛选特征点;最后,采用灰度质心法,实现特征点的旋转不变性。实验结果表明,相较于ORB-SLAM2中的特征提取方法,本文提出的QOARB算法在保证特征点均匀程度的同时,特征提取速率提高了10.65%,匹配正确率和正确匹配数分别提升1.01%和6%。

Feature extraction rate and homogenization have an important impact on the performance of visual SLAM.In order to improve the feature extraction rate of traditional visual SLAM system,the uniform extraction of image features is realized,an adaptive threshold based AGAST feature homogenization extraction algorithm(QOARB)is proposed.Firstly,scale invariance of feature points is realized by constructing image pyramid;Secondly,according to the gray value and variance of the image,the initial extraction threshold of AGAST corner is calculated to improve the feature extraction rate of the algorithm on different images;Then,the improved quadtree algorithm is used to uniformly screen the feature points;Finally,the graysacle centroid method is used to realize the rotation invariance of feature points.The results show that compared with the feature extraction method in ORB-SLAM2,the QORAB algorithm proposed in this paper not only ensures the uniform extraction of feature points,but also increases the extraction rate by 10.65%,the matching accuracy and the number of correct matches by 1.01%and 6%,respectively.