弱标定立体图像对的欧氏极线校正框架  

Euclidean epipolar rectification frame of weakly calibrated stereo pairs

作　　者：李国栋[1,2] 田国会[2] 王洪君[1] 尹建芹[3] 

机构地区：[1]山东大学信息科学与工程学院,山东济南250100 [2]山东大学控制科学与工程学院,山东济南250061 [3]济南大学信息科学与工程学院,山东济南250022

出　　处：《光学精密工程》2014年第7期1955-1961,共7页Optics and Precision Engineering

基　　金：国家863高技术研究发展计划资助项目(No.2006AA040206;No.2009AA04Z220);国家自然科学基金资助项目(No.61075092;61203341);山东省自然科学基金资助项目(No.ZR2011FM011);山东大学自主创新基金资助项目(No.2011JC017)

摘　　要：为了快速且直观地对相机弱标定(即相机内参数已知而外参数未知)情况下的立体图像对进行欧氏极线校正,提出了一套完整的欧氏极线校正框架。首先,针对两视图对应相机的位姿信息估计问题,提出一种新的不依赖矩阵奇异值分解运算的本质矩阵分解和唯一解确定算法。然后,对传统欧氏极线校正问题进行推广,并利用估计出的相机位姿构造出具有明确几何意义的无穷单应变换对,进而将其施加到立体图像对来完成欧氏极线校正。最后,利用8组SYNTIM立体图像对验证了所提出校正框架的正确性和校正精度。实验结果表明:该欧氏极线校正框架不仅形式直观,计算简单,易于实现,而且具有明确的几何意义。与其他欧氏极线校正方法比较,提出的方法能够获得更高的校正精度,且更好地消除了立体图像对的水平或垂直视差。To quickly and intuitively finish the Euclidean epipolar rectification of weakly calibrated stereo pairs（namely,the camera intrinsic parameters are known and the camera external parameters are unknown）,an integral Euclidean epipolar rectification frame was proposed.Firstly,a novel essential matrix decomposition and unique solution determination algorithm independent on the Singular Value Decomposition（SVD）operation was proposed to determine the position and orientation information between two cameras.Then,the problem of traditional Euclidean epipolar rectification was generalized,and the infinite homography matrixes were constructed by the estimatedcamera position and orientation to finish the rectification process of stereo pairs.Finally,the eight groups of SYNTIM images were used to validate the correctness and accuracy of the rectification frame.Experimental results indicate that the proposed frame has the advantages of intuitive form,low computational complexity and easy implementation,as well as the explicit geometrical meanings.As comparison with traditional Euclidean epipolar rectification method,it can well remove the horizontal or vertical disparities of stereo pairs,and can offer higher rectification accuracy.

关 键 词：相机标定 对极几何 极线校正 立体图像对 本质矩阵分解 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]