自适应阈值的基础矩阵估计算法  

Robust estimation method of fundamental matrix with a self-adaptive threshold

作　　者：吴宇豪 曹雪峰[1] 安籽鹏 WU Yuhao;CAO Xuefeng;AN Zipeng(Information Engineering University,Zhengzhou 450000,China)

机构地区：[1]信息工程大学

出　　处：《测绘科学》2019年第11期22-27,34,共7页Science of Surveying and Mapping

基　　金：信息工程大学优秀青年基金课题项目(2105010205);地理信息工程国家重点实验室开放基金项目(SKLGIE2016-M-3-4);国防科技项目(3601015)

摘　　要：针对RANSAC算法在估计基础矩阵过程中需要人为设定阈值的问题,该文提出一种自适应阈值的基础矩阵估计算法。该算法首先引入ORSA算法,计算内点集以及基础矩阵,随后将得到的内点集与基础矩阵作为最小中值算法的初始值做进一步加权优化,在保证基础矩阵估计精度的前提下得到更好的内点集。其中,利用ORSA算法估计时通过计算误匹配警报数(NFA)值评判估计精度,舍去了RANSAC算法中人为设定阈值的步骤;利用最小中值算法加权优化的过程中采用最小化误差中值的方式,避免人为设定阈值。实验结果显示,该算法在保证基础矩阵估计精度的同时,能够获得最佳的内点集,且具有一定的抗噪声能力。In order to solve the problem that the RANSAC algorithm needed to set an empirical threshold to estimate the fundamental matrix,an adaptive threshold-based matrix robust estimation method that combined the ORSA algorithm and the LMedS algorithm was proposed.This method first used the ORSA algorithm to obtain the interior point set,and then used the LMedS algorithm to further optimize the weighted optimization on the interior point set.To a greater extent,a better inner point set was obtained on the premise of ensuring the accuracy of the base matrix estimation.Among them,when using the ORSA algorithm to calculate the interior point set,the accuracy of the estimation was evaluated by calculating the number of false alarms(NFA),the threshold was determined,and the artificially set thresholds in the RANSAC algorithm were discarded;the weighted optimization was performed using the LMedS algorithm.In the process,artificially setting the threshold was avoided by minimizing the median error.Simulation experiments and actual photo experiments showed that this method could obtain the best inner point set while guaranteeing the accuracy of the fundamental matrix estimation,and had a certain anti-noise ability.

关 键 词：基础矩阵 自适应阈值 对极几何 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]