基于质点系运动原理的像机位姿估计算法  

Camera Pose Estimation Algorithm Based on Particle System Kinematics

作　　者：陈鹏[1] 胡广大[1] 闫桂荣[2] 

机构地区：[1]北京科技大学自动化学院,北京100083 [2]美国德州大学艾尔帕索分校土木工程系,美国艾尔帕索tx79968

出　　处：《光学学报》2014年第1期121-128,共8页Acta Optica Sinica

摘　　要：大多数基于单幅二维(2D)图像的像机位姿估计算法都是在已知全部或者部分三维(3D)/2D特征点对应关系的基础上设计的,而对于3D/2D特征点对应关系完全未知的情况则很少涉及。利用经典力学中的质点系运动原理设计了一种像机位姿估计算法,该算法能够在未知特征点对应关系的情况下,同时确定3D/2D特征点对应关系与像机位姿。此外,利用空间共线性误差和匹配矩阵,所提出的算法不仅能够处理3D/2D特征点一一对应的情况,而且能够处理部分3D特征点被遮挡,以及图像中存在2D伪特征点的情况。通过实验,并与其他相关算法进行比较,结果表明所提出的算法能够在不增加计算复杂度的条件下有效地匹配特征点并估计像机位姿,而且受图像噪声和2D伪特征点的影响较小。Most pose estimation algorithms using a single two-dimensional （2D） image are designed based on totally or partially knowing the correspondence between three-dimensional （3D） and 2D feature points, but few have involved in a correspondenceless case. A camera pose estimation algorithm is proposed by using particle system kinematics in classical mechanics. It can simultaneously decide the correspondence and the camera pose. Besides, by introducing object space collinear error and matching matrix, the proposed algorithm can be used not only in the case when the correspondences of 3D/2D feature points are one to one, but also in the case when the 3D feature points are partially occluded and when there are 2D false feature points in an image. Through experiments and comparison with other algorithms, the result shows that the proposed algorithm can effectively find the correct correspondence and estimate camera pose without increasing computational complexity, and the impact of image noises and 2D false feature points on the algorithm is little.

关 键 词：测量 位姿估计 质点系 无对应 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]