离散型概率分布的ORB图像特征点误匹配剔除算法

Mismatch Elimination Algorithm Based on Discrete Probability Distribution of ORB Feature Points

针对图像特征点产生的误匹配问题,提出了根据特征匹配点的欧氏距离的概率分布的筛选和剔除方法。首先,从视频流中提取相邻的两张图像帧,基于ORB(Oriented Fast and Rotated Brief)特征法分别提取特征点并初始匹配;然后,利用特征点的像素坐标计算出匹配点对的欧氏距离;再用概率分布列描述这组离散化的距离值;最后,选取最大概率作为阈值模型,从而剔除错误匹配。实验数据显示,该算法与经典的随机抽样一致性算法相比(RANSAC),剔除误匹配后的正确匹配数相当,但运行时间大幅度降低,不仅保证了特征点的匹配精度,也提高了算法实时性。

In order to solve the problem of mismatching caused by the feature point of images,a screening and culling method based on the probability distribution of Euclidean distance of feature matching points is proposed.First,two adjacent image frames are extracted from the video stream,and feature points are extracted and initially matched based on the ORB feature method. Then,the pixel coordinates of the feature points are used to calculate the Euclidean distance of the matching point pair. This set of discrete distance values is described by the probability distribution column. Finally,the maximum probability is selected as the threshold model to eliminate false matches. The experimental data shows that the number of correct matches after the new algorithm eliminates mismatches is equivalent to that of Random sample consensus(RANSAC) algorithm,but the running time is greatly reduced. It is not only guarantees the matching accuracy of the feature points,but also improves the real-time performance of the algorithm.