摘要： 提出一种无相机标定的立体图像对的极线校正新方法。该校正方法并不依赖基本矩阵F的精确求解，而是通过空间变换法分析校正前后图像点对应关系，依此分解并参数化描述极线变换矩阵，直接利用极线方程和图像的对应点集建立误差平方和函数，并运用非线性最小二乘法求解，使该函数取得最小值的变换参数。实验证明，该校正方法能够较好地消除垂直视差，图像产生的变形较小。

关键词: 极线几何, 基础矩阵, 立体图像对, 投影几何

Abstract: A new image rectification method deals with the problem of epipolar rectification in the uncalibrated case is proposed. It is implemented without explicit computation of the epipolar geometry or fundamental matrix. It decomposes the transformation matrix into more simple form and using least-squares method to solve the function of sum of square errors based on the set of correspondence points. Related experiments show that the proposed method can rectify the image pairs fast and accurately, and the image distortion induced is little.

Key words: epipolar geometry, fundamental matrix, stereo image pairs, projection geometry

中图分类号: 

• TP393