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.
出处
《光电工程》
CAS
CSCD
北大核心
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)
关键词
本质矩阵
三维测量
相对定向
五点算法
反投影残差
essential matrix
3D measuring
relative orientation
five-point algorithm
re-projection residual