基于加权网格运动统计的改进ORB算法

Improved ORB algorithm based on weighted grid motion statistics

针对传统ORB算法存在特征点分布集中,在光照条件变化的情况下特征点提取不稳定,匹配准确率下降的问题,提出一种改进ORB算法.首先构建高斯差分金字塔,然后将各层图像划分成多个区域,根据区域内像素值的中位数计算每个区域的阈值,提取每个区域的特征点,之后采用双向匹配算法对特征点进行粗匹配,最后提出一种加权网格运动统计(Weighted Grid Motion Statistics,WGMS)算法对粗匹配结果进行优化.实验结果表明改进ORB算法可以使特征点提取更为均匀,在光照变化的情况下,可以稳定地提取特征点,算法耗时较传统ORB算法慢了10.7%,匹配准确率提高了41.7%.

Aiming at the problems of traditional orb algorithm,such as the distribution of feature points is concentrated,the feature point extraction is unstable and the matching accuracy is decreased when the illumination condition changes,an improved orb algorithm is proposed.Firstly,the Difference of Gaussian is constructed,and then each layer image is divided into several regions.According to the median value of pixels in the region,the threshold value of each region is calculated,and the feature points of each region are extracted.Then,the feature points are roughly matched by bidirectional matching algorithm.Finally,a weighted mesh motion statistics algorithm is proposed to optimize the coarse matching results.The experimental results show that the improved ORB algorithm can make the feature points extraction more uniform,and it can extract feature points stably under the condition of changing illumination.The time-consuming of the algorithm is 10.7%slower than the traditional ORB algorithm,and the matching accuracy is improved by 41.7%.