基于对极几何约束的景象匹配研究

Study on Scene Matching Based on Epipolar Geometric Constraint

提出了一种图像配准方法来解决实时图与基准图空间不对准问题。它是利用随机采样算法估计基本矩阵 ,恢复实时图与基准图之间对极几何 ,然后基于对极几何约束 ,剔除误匹配点 ,得到精确匹配控制点 ,计算出全局仿射变换 ,从而对实时图进行校正。该方法的特点是精确、稳定和全自动。采用真实图像实验结果表明 ,该方法是行之有效的。

Scene matching is very important for end-control-and-guide precision. An image registration method is presented to solve the displacement between real time and reference images. The method uses a random sampling algorithm to estimate the fundamental matrix and to recover the epipolar geometry in the image pair. A correlation technique is used to find an initial set of matches and a robust technique is used to discard false matches in this set, so that accurate matches are eventually found by using the recovered epipolar geometry. The global affine transformation is obtained from final set of points that aligns correctly the image pair. The method has characteristics of the accuracy and stability of estimation, and the automated solution. Experimental results with real images show that the method is effective.