一种基于虚拟立体靶标的线阵相机标定方法  

A line-scan camera calibration method based on virtual stereo pattern

作　　者：王子辰[1,2,3] 戴明[1,3] 汪永阳[1,2,3] 李刚[1,3] 李家德[1,2,3] 郎小龙[3] 

机构地区：[1]中国科学院航空光学成像与测量重点实验室,吉林长春130033 [2]中国科学院大学,北京100039 [3]中国科学院长春光学精密机械与物理研究所,吉林长春130033

出　　处：《光电子．激光》2013年第5期867-873,共7页Journal of Optoelectronics·Laser

基　　金：国家重点基础研究"973计划"(2009CB7240012A);装备预先研究(402040203);中国科学院航空光学成像与测量重点实验室开放基金(Y2HC1SR123)资助项目

摘　　要：针对线阵相机标定的过程中特征点容易丢失的事实,提出了一种线阵相机标定方法。首先,利用设计的平面靶标与相机多次曝光巧妙地构建了虚拟立体标定靶标,通过分析与数学建模得到靶标上标定特征点空间位置与像点的对应关系;其次,使用最小均方估计方法对特征像点中心进行亚像素估计,并使用基于最小二乘迭代方法对多特征点超定方程进行求解,讨论了标定过程中相机与靶标非平行时对标定结果的影响;最后,使用提出的方法对线阵相机标定进行实验验证。理论分析与实验结果表明,本文的线阵相机标定方法简单灵活,标定特征点个数不拘泥于靶标的制作,相机与靶标的位置无需严格平行,平均标定精度达0.26pixel,具有较好的实用价值。This paper proposes a novel method that can be used for improving line-scan camera calibration according to the fact that calibration points are easy-missing during calibration. First, we use the specially designed planar pattern and multiple exposures of line-scan camera to produce a virtual stereo calibration pattern. We then analyze the characteristics of virtual stereo calibration pattern and established a mathematical model to seize the relation between the coordinates of calibration points on the pattern and those of their images. Second, we employ the technique of least mean square to estimate the center of the calibration image points in sub-pixel resolution. The over-determined equations for multiple feature points are thus able to be solved by iterative least squares. We also emphatically analyze the impact on the calibration result that is caused by the non-parallel nature between the camera and calibration pattern. Finally, both the analysis and experimental results reveal that this proposed line-scan camera calibration method is not only convenient and flexible,but the calibration points are also independent of calibration pat- tern, and the condition of parallelism between the calibration pattern and camera is not required. All these features make our devised line-scan camera calibration method accurate, as precise as 0. 26 pixel, and more practical.

关 键 词：线阵相机标定 虚拟立体靶标 最小均方估计 最小二乘迭代 标定误差 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]