本质矩阵五点算法伪解的两种剔除策略  

Two Removal Tactics of Pseudo Solutions for Essential Matrix Five-point Algorithm

作　　者：王文斌[1] 刘桂华[1] 刘先勇[1,2] 邱志强 

机构地区：[1]西南科技大学信息工程学院,四川绵阳621010 [2]绵阳铁牛科技有限公司,四川绵阳621010

出　　处：《光电工程》2010年第8期46-52,共7页Opto-Electronic Engineering

基　　金：四川省科技厅国际合作项目(2009HH0023)

摘　　要：本质矩阵五点算法是实现三维测量中双视图相对定向的常见方法,在其计算过程中常常采用多项式求解技术,从而引发了解的多异性。为了确定正确解,提出了五点算法的两种改进实现形式,用于消除多异解。它首先用点在相机前的约束排除非物理可实现解,然后在剩余的可能解中分别计算当前双视图中所有公共点的Sampson距离或反投影残差之和,最小值对应的相机参数即为正确的定向参数值。仿真和真实实验均证明了两种策略的可行性和正确性,且基于Sampson的方法较基于反投影的方法速度快。The five-point algorithm of essential matrix is a common way to achieve relative orientation of the two-view images in 3D measuring.Polynomial solving techniques,which lead to polysemia while computing,are always adopted during the computing process.In order to determine the right solution,two improved methods for five-point algorithm are proposed to avoid multi-solutions.First of all,the inconsistent solutions of physical model were excluded with cheirality constraint.Secondly,the rest error solutions can be solved by computing sums of Sampson distance of all the common points or re-projection residual.In the two-view images,the minimum value among sums is just the correct orientation parameter values.Both simulation and real images experiments have proved the feasibility and correctness of the two tactics.In most cases,methods based on Sampson are much quicker than that based on re-projection.

关 键 词：本质矩阵 三维测量 相对定向 五点算法 反投影残差 

分 类 号：TP391.7[自动化与计算机技术—计算机应用技术]