摄像机标定的棋盘格模板的改进和自动识别  

Checker pattern improvement and fully-automatic identification for camera calibration

作　　者：张浩鹏 王宗义[1] 吴攀超[1] 林欣堂[1] 

机构地区：[1]哈尔滨工程大学自动化学院,哈尔滨150001

出　　处：《仪器仪表学报》2012年第5期1102-1109,共8页Chinese Journal of Scientific Instrument

基　　金：黑龙江省自然科学基金(AF200921)资助项目

摘　　要：为了克服在摄像机标定过程中需要使用者给出标定模板的附加信息,或全自动标定点识别算法在遮挡、不均匀照明、大视角和摄像机镜头畸变情况下不能检测出标定点的缺点,提出一种改进的基于基准点标记的棋盘格模板以及相应的全自动识别算法。新的摄像机标定模板以基准点标记代替传统棋盘格的黑白方块,从而使全自动识别算法识别出标记的位置。利用模板中标记按照标记ID从小到大的顺序排列的先验知识,估计丢失的标定点位置。为了提高丢失标定点在图像中初始位置的估计,算法估计径向畸变参数,从而克服了畸变对识别的影响。为了提高标定点的定位精度,利用高精度的鞍点检测器,从而标定点的定位精度小于0.05像素。为了检测鞍点的有效性,算法提出2种滤波准则,最终得到有效的标定点。识别算法是有效的且不需要任何参数。实验结果表明,对于同样的摄像机和背景,使用改进的棋盘格模板及其识别算法获得的标定点进行摄像机标定的投影误差比ARTag减少70%。In order to overcome the shortcomings that in camera calibration the user needs to give additional information of calibration pattern or fully-automatic identification algorithm of calibration points can not detect calibration points under the conditions of significant occlusions,uneven illumination,observation with extremely viewing angles and lens distortion,an improved checker pattern based on fiducial markers is designed,and the corresponding fully-automatic identification algorithm of the calibration points is proposed.The new camera calibration pattern replaces the black and white squares in traditional checker pattern with the fiducial markers,so the fully-automatic identification algorithm can locate the positions of the markers.Using the priori knowledge that the markers are arranged sequentially in the calibration pattern according to the marker ID from small to large,the missed calibration points can be located.In order to improve the estimate of the initial positions of the missed points in the image,the algorithm estimates the radial distortion parameters,so the identification algorithm overcomes the impact of lens distortion on identification.In order to improve location accuracy of calibration points,a subpixel-accuracy saddle point finder is used,so the location accuracy of calibration points is less than 0.05 pixels.In order to validate the potential points,two filtering criteria are proposed,and finally the valid calibration points are obtained.The detection algorithm is efficient and does not need any parameters.Experiment results show that after calibration the reprojection errors are 70% lower than that of the state-of-the-art ARTag using the calibration points obtained with the improved checker pattern and identification algorithm for the same camera and background.

关 键 词：标定点检测 全自动识别 棋盘格模板 摄像机标定 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]