一种联合直线特征的基础矩阵计算方法  

Method for Fundamental Matrix Estimation Combined with Line Features

作　　者：周凡[1] 邵世维[2] 吴建华[3] 付仲良[1] 

机构地区：[1]武汉大学遥感信息工程学院,湖北武汉430079 [2]武汉市国土资源和规划信息中心,湖北武汉430014 [3]江西师范大学地理与环境学院,江西南昌330022

出　　处：《光学学报》2013年第10期188-195,共8页Acta Optica Sinica

基　　金：国家自然科学青年基金(41201409)

摘　　要：为了解决缺乏明显特征点给基础矩阵计算带来的难题,针对水平直线与垂直直线所特有的极几何关系,提出了一种可以联合这两种直线特征的基础矩阵计算方法。首先以相机方位元素、相对姿态为基础矩阵的参数化要素,构建基于水平直线与垂直直线的等量方程式,然后以特征直线作为RANSAC算法的内点筛选出特征点的最优子集,再结合M-estimators算法确定各个要素的权值,最后构建统一的平差模型进行计算。模拟数据与真实图像的对比实验结果表明,该方法切实可行,能够减少传统基础矩阵计算方法对特征点的依赖,特征直线的引入也能在一定程度上提高计算结果的精确度与稳定性。In order to resolve the current problems of fundamental matrix estimation caused by lacking of obvious feature points, according to the epipolar geometry of the horizontal Lines and vertical lines, a new method of calculating the fundamental matrix that can be combined with these two kinds of line features is proposed. Firstly, camera orientation elements and relative attitude are used as the elements of fundamental matrix, the equivalent relationship based on horizontal lines and vertical lines is deduced, and then the line features are made as the interior point of the RANSAC algorithm, after that the weight of each factor is determined combined with M-estimators, and finally the unified adjustment model is built. Experimental results of simulation data and real images show that the proposed approach is feasible in practice, and can reduce the dependence of intraditional method on feature points, and the introduction of the line features can improve the accuracy and stability of the calculation results to a certain extent.

关 键 词：机器视觉 基础矩阵 对极几何 直线特征 稳健性 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]