本质矩阵优化分解的相对位姿估计  

Relative Pose Estimation Based on Improved Essential Matrix Decomposition

作　　者：张振杰[1,2] 郝向阳 孙国鹏[1,2] 程传奇 

机构地区：[1]信息工程大学,河南郑州450001 [2]北斗导航应用技术河南省协同创新中心,河南郑州450000

出　　处：《测绘科学技术学报》2017年第4期353-357,共5页Journal of Geomatics Science and Technology

摘　　要：针对从影像恢复摄像机相对位姿的问题,提出了一种基于李群表示的本质矩阵快速分解的位姿估计算法。通过加权最小二乘方法优化了本质矩阵;利用本质矩阵和平移向量的关系求出了平移向量;由本质矩阵和位姿参数的等式关系建立目标函数,基于姿态的李群表示推导了旋转矩阵迭代估计过程;优化了唯一解确定的约束条件,避免了特征点的三维重建。仿真实验和真实图像实验表明提出的算法精度和鲁棒性均优于传统算法,算法效率得到明显提高。提出的算法避免了矩阵奇异值分解运算和大量的矩阵计算,而且只需对两组解进行唯一解确定,能够实现相对位姿的快速高精度估计。In order to solve the relative camera pose estimation from images, a novel fast relative pose estimation method based on improved essential matrix decomposition is proposed. Essential matrix is optimized by weighted least squares. The translation vector can be calculated by the relation between essential matrix and itself. The objective function based on the equality between essential matrix and pose parameters is built, and rotation matrix is iteratively computed based on orientation of Lie group representation. Improved unique solution constraint avoids reconstruction of features. Experimental results indicate that proposed method has higher precision and better robust than traditional algorithm, and the computing efficiency is obviously improved. The proposed algorithm avoids singular value decomposition operation and lots of matrix operations, and just needs to determine the unique solution from two solutions. The proposed algorithm can realize the relative pose estimation with high precision and fast computation.

关 键 词：相对位姿估计 本质矩阵 目标函数 迭代计算 李群表示 

分 类 号：P231[天文地球—摄影测量与遥感] TP391.4[天文地球—测绘科学与技术]