基于RANSAC的单应性矩阵估计优化算法  

Homography matrix estimation optimization algorithm based on RANSAC

作　　者：赖焕杰 孟祥印[1] 肖世德[1] 胡锴沣 李召鑫 LAI Huanjie;MENG Xiangyin;XIAO Shide;HU Kaifeng;LI Zhaoxin(School of Mechanical Engineering,Southwest Jiaotong University,Chengdu 610031,China)

机构地区：[1]西南交通大学机械工程学院,四川成都610031

出　　处：《传感器与微系统》2023年第8期135-138,共4页Transducer and Microsystem Technologies

摘　　要：图像特征匹配是机器视觉处理技术的关键环节。在图像特征匹配中,需要根据检测到的特征点及其相应的特征描述子进行特征点匹配。在众多匹配方法中,传统随机抽样一致性(RANSAC)算法因为能利用随机抽样从样本集中剔除误匹配点,再对单应性矩阵进行估计,而获得了广泛的应用,但其模型参数估计依然存在精度不高和效率较低的问题。本文用基于先验概率抽样的方法代替随机抽样方法,同时,使用二次匹配代替正向匹配来计算重投影误差,使用标准测试图集进行实验。实验结果表明:单应性矩阵的估计精度和效率分别提升了48.42%和53.57%。Image feature matching is the key link of machine vision processing technology.In image feature matching,it is necessary to match the feature points according to the detected feature points and their corresponding feature descriptors.Among the many matching methods,the traditional random sampling consensus(RANSAC)algorithm has been widely used because it can use random sampling to eliminate mismatching points from the sample set and then estimate the homography matrix,but its model parameter estimation still has the problems of low precision and low efficiency.The method based on prior probability sampling is used instead of random sampling method,at the same time,quadratic matching is used instead of forward matching to calculate the reprojection error,and the standard test atlas is used for experiments.The experimental results show that the estimation precision and efficiency of the homography matrix are improved by 48.42%and 53.57%,respectively.

关 键 词：特征匹配 单应性矩阵估计 随机抽样一致性算法 先验概率 

分 类 号：TP294.2[自动化与计算机技术—检测技术与自动化装置]