未校准相机最小化重采样失真对极线校正方法  

Minimization of resampling distortion in epipolar rectification for uncalibrated cameras

作　　者：武国梁[1] 李素梅[1] 杨建[1] 刘畅[1] 

机构地区：[1]天津大学电子信息工程学院,天津300072

出　　处：《信息技术》2013年第4期103-106,109,共5页Information Technology

摘　　要：校正作为立体视觉的先决条件,通常会产生不必要的几何变形。文中提出一种新的针对未校准相机的立体图像校正方法以减少几何失真。在图像级上限制相机的单应性矩阵接近刚性变换,并引入一个惩罚项以进一步减少在像素级的失真。在不显示计算基本矩阵的前提下整个图像校正问题构成了一个函数能量最小化的问题。实验结果表明该方法实现了最小化重采样失真的校正。Epipolar rectification, a prerequisite for stereo vision, usually a^ses unwanted geometric deformation. This paper proposes a new rectification method for uncalibrated stereo images to reduce geometric distortion. Both homographies are restrained to close to rigid ones in the image level and introduce a penalty term to further reduce the distortion in the pixel level. The image rectification problem is formulated into the minimization of an energy function without explicit computation of fundamental matrix. The experiments show that the proposed method achieves promising rectification performance while introduces less geometric distortion.

关 键 词：校正 单应矩阵 雅可比行列式 图像失真 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]