一种基于双目立体视觉的图像校正算法设计

立体校正作为立体匹配的先决条件,能消除图像对对应像素的垂直视差,有效地提高立体匹配的精确度.针对校正带来的图像失真问题,研究了一种新的非标定相机的立体校正方法.该方法首先利用SIFT算法提取匹配点,再结合极线校正准则和雅可比行列式来构建能量函数,通过最小化能量函数求得基础矩阵和校正投影变换矩阵,并在像素级上优化了投影变换矩阵.研究结果表明:所设计的校正方法有效地减小了校正误差,更好地保持了原图像的重采样,降低了图像的失真度.

Design of an Image Rectification Algorithm Based on Binocular Stereo Vision

A precondition for stereo matching is stereo rectification which can eliminate the corresponding pixels vertical parallax of image pair and improve the precision of stereo matching effectively.In order to elminate image distortion caused by rectification,a new rectification method for uncalibrated cameras is researched.SIFT algorithm is used to extract corresponding points,and combinating polar rectification criterion and Jacobian determinant to build energy function which is minimized to obtain fundamental matrix and projective transformation.And then,the projection transformation matrix is optimized on pixel level.The research results show that the rectification decreases the rectification error,keeps the resampling of original image better,and reduces the image distortion.