基于三维参照系的摄像机分步定标

Strategic calibration method in 3D reference system

基于三维参照系的定标方法,是一种传统的定标方法,需要三维控制场来提供足够的数据信息,定标精度较高,但对图像噪声不够鲁棒,并且非线性算法受出格点影响很大,甚至有时出现优化过程不收敛。本文提出了一种基于三维参照系的摄像机分步定标方法,利用对极几何关系对图像间的特征点进行分类,以剔除出格点,再选择合适的方法进行定标,实验表明分步定标法较好解决了原有技术对噪声敏感和非线性算法不收敛的情况,提高了各种算法的估计精度,并且三维重建误差也减少了30%左右。

In this paper, we address a new strategic calibration method in 3D reference system. Epipolar geometry constraint is applied to classify the features points between images; then the outliers, which disturbed by image noise or bad locations, are deleted; finally linear or non-linear algorithms are employed to compute the camera parameters. Experimental results show that our approach is robust to noise and improves the convergence of non-linear algorithms.