双目立体视觉是光学被动三维测量的主要方法。为了快速、准确地寻找对应点,通常要将立体图像对进行校正,消除垂直视差。提出了一种无需标定相机的立体图像对校正方法。该方法从基本矩阵计算初始透视投影、旋转和竖直平移变换矩阵,然后以对应点坐标为基础对这些变换矩阵进行优化计算,从而有效地避免了优化计算的局部最小值,而且不过分依赖基本矩阵的计算精度。通过对提出的方法和完全基于基本矩阵的图像校正方法以及无需基本矩阵计算的图像校正方法进行实验比较,结果表明提出的方法图像校正速度快,能有效地消除垂直视差,而产生的图像变形较小。

The binocular stereovision is a primary approach to realize optical passive 3-D measurement. To search corresponding points fast and accurately, the stereo pairs must be rectified. A method is presented to rectify stereo pairs without calibration for cameras. In this method the initial values of perspective projective matrix; rotative matrix and translative matrix are calculated from the fundamental matrix, than these matrices are optimized based on the coordinate value of corresponding points. This method can avoid the local least value in the optimal calculation effectively, and the dependence on the calculation precision of the fundamental matrix excessively. By the experiment of image rectification to compare this method put forward this here with two typical methods that one depend on fundamental matrix calculation and the other dispenses with fundamental matrix calculation, it is shown that this method can rectify stereo pairs fast and accurately, but bring little image distortion.